* Legacy MBQL documentation

* Do not commit dox

* Subject the legacy MBQL stuff to the same stricter linting rules as MLv2

* PR feedback

New endpoint accepts entity and entity ids in form of `database=1&dashboard=2&question=3`. If you don't supply
`&include=overrides`, then it tries to find configs directly referencing supplied entities and updates their
`invalidated_at`. If you supply `&include=overrides`, all the referenced cards are
updated (`report_card.cache_invalidated_at`).

Cache strategies then select the maximum `invalidated_at` in their logic. EE-only.

resolves #40548

* Wow

* Test fix 

* Fixes

* Actions should use strings for column names (fix :update-row and :create-row normalization)

* MLv2 schema should check against keys for the other query type

* Ok, have I fixed things?

* More fixes 

* Fix indentation

* Another round of test fixes. 

* Hopefully the last few test fixes 

* We need to test normalization for queries that have keyword keys as well.

* Fix Cljs i18n namespaces

* Sort namespaces

* Only test against H2

* Rename `metabase.mbql` to `metabase.legacy-mbql`

* Fix Kondo warnings

* Test fixes 

* Register MBQL clause schemas and test fixes 

* Test fixes and PR feedback

* Test fix

* Remove the normalization tests

* Test fixes 

* Fix kondo

* Fix import

* Another fix 

* Merge

* FIXES

* Add another missing REQUIRE

* first move to a db api

* mdb.spec into mdb

* mdb.u/isa -> mdb.query/isa

* mdb.u/qualify -> mdb.query/qualify

* last of mbql.u -> mbql.query and mbql.u

* remove last of the mdb.connection/db-type and use from mdb/db-type

* remove the last non-cmd usages of mdb.connection

had some circularity in

models.interface requires mdb/db-type for sql flavored "now"
mdb requires db/setup
which requires db/custom_migrations to ensure they are on cp
custom migration requires json functions in models.interface

so had to do a dynamic require for the db-type

* get db-type from mdb not mdb.env

* invoke `(mdb/app-db)` and not pass a var. lots of things fail

* mdb.setup/migrate! into mdb with potemkin

* remove get-connection. call .getConnection on the app-db

perhaps invites clumsy usage outside of `with-open` if it's a top level
function

* namespace docstrings

* last few and grab db.env/db-file into api ns

* lint cleanup

* addresses PR comments

* Remove references to shared/ directory which no longer exists

* Update shadow-cljs and pinned cider/nrepl versions

* Support legacy metadata and raw Dates for brush filter helper functions

* Fix fn name

* Apply suggestions from code review

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

---------

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

Makes /api/properties and /api/premium-features/token/status return valid structures with information about invalid token, plus logs the error about token being invalid in logs. Also this information is used to show a PaymentBanner on top of the screen so that admins can understand what is going on.

Resolves #33858

* Release script fix

* Add test 

* Metabase lib PoC RFC [ci skip]

* Cleanup [ci skip]

* `:lib/uuid` overhaul [ci skip]

* More tests [ci skip]

* Basic join support [ci skip]

* Simplify a few things, more tests, etc. [ci skip]

* Pipeline queries FTW [ci skip]

* Metabase® Lib™ v2

* Fix cljs test assertions

* No bees please

* Update docstring

* Mega simplification

* Fix Kondo warnings

* Address PR feedback

* Fix Malli util for Cljc

* Oops, `instrument!` should be public

* Cloverage do not instrument JS util namespace

* Fix Cloverage again

* Don't count MySQL or Postgres namespaces for Coverage

* Address PR feedback

* Move dispatch value keyword function to util

* Revert acccidental require

* Test fix

* Fix Cljs =?

* Fix totally unrelated broken links

* Delay substituting persisted native query until after analysis

We were substituting persisted queries into the query at too early a
stage.

Card 729 is a persisted model of `select * from categories`

```clojure
{:type :query,
 :query {:source-table "card__729",
         :expressions {:adjective ["case"
                                   [[[">" ["field" 100 nil] 30]
                                     "expensive"]
                                    [[">" ["field" 100 nil] 20]
                                     "not too bad"]]
                                   {:default "its ok"}]},
         :aggregation [["count"]],
         :breakout [["expression" "adjective" nil]
                    ["field" 101 nil]]},
 :database 3,
 :parameters []}
```

This would immediately get turned into an equivalent

```clojure
{:type :query,
 :query {:source-table 12
         :persisted-info/native "select \"id\", \"ean\...

insert next-gen permission paths alongside currently active permission paths + classification (#27911)

* implement move, which returns v2 paths

- TODO: insert these into the db

(move v1-path) => [v2 paths]

* cleanup + add some schemas

* generative tests 4 permission path classification

* whitespace lint

* detect data, query, and paths for v2

* calling move on v2 paths is a no-op

* differentiate between v1 and v2 permissions

quickchecking for move, classify-path, and classify-data-path

* fix tests + add idempotency test

* add tests for classification of permission paths

- rename move to ->v2-path
- move some fxns around
- ascii art in test

* making the legos line up

- need to insert both v1 and v2 versions of paths (of course)
- valid-path? has to allow v2 paths to be inserted

* replace mu/with-api-error-message

* linter + code quality fixes

* privatize rx->kind

* remove some changes that should be 4 anotherbranch

* revert ns

* delete v2 permissions in

- they aren't handled by the perm graph parser, so they don't get propagated into "old graph", so the diff between old and new indicates that they need to be rewritten.

* only delete v2 paths for the current group_id

- reorder declarations in models.permissions

* remove extra line break

* Update src/metabase/models/permissions.clj

fix typo

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

---------

Co-authored-by: metamben <103100869+metamben@users.noreply.github.com>

* Switch to Toucan 2 part 1

* Fix Kondo errors

* Switch to Honey SQL 2, Part 1

* Test fix

* Fix Kondo error

* [Toucan 2 Prep] Add helper for extending `IModel`; namespace Toucan properties

* Clean namespaces

* Test fixes 

* [Toucan 2 Prep] Add `define-*-hydration-method` macros

* Sort namespaces

* Disable Eastwood `:unused-private-vars` linter

* [Toucan 2 Prep] Add helper for extending `IModel`; namespace Toucan properties

* Clean namespaces

* Test fixes 

The new macro is called defendpoint. The previous defendpoint is now defendpoint-schema, as per @camsaul's idea. It'll make it easier to realize that we should use the Malli version of defendpoint going forward.

* adds malli defendpoint + linting + tests

- I decided to add a new defendpoint macro -- it is nearly exactly the
  same as `defendpoint`. If the duplication is an issue I can handle that too.

* malli-defendpoint -> defendpoint-malli

* defendpoint -> defendpoint-schema

* defendpoint-malli -> defendpoint

* fine tuning error messages + comments

- fix kondo config

* put back dox-for-plumatic

* fix name collision

* adds local malli description ns + test

* update alias in test

* linting fix

* linting fix pt. 2

* hook up umd/describe

* ≈ [WIP] [ci skip]

* Minor cleanup [ci skip]

* Code cleanup. [ci skip]

* Kondo in CI should fetch library configs

* Bump Methodical version

* Change name to `=?`

* Add `#exactly` reader tag

* Add `#schema` data reader

* Fix dev deps indentation and add `algo.generic`

* Add `approx=`

* Improved version of `#approx`

* Just check in third party Kondo config for now instead of fighting CI

* Update test/metabase/test_runner/assert_exprs/approximately_equal_test.clj

Co-authored-by: Tim Macdonald <tim@metabase.com>

* Address PR feedback and fix sequence comparison

Co-authored-by: Tim Macdonald <tim@metabase.com>

* [Toucan 2 prep] Don't invoke Toucan models as functions

* Some fixes

* Test fixes

* Fix some small things

* Add Kondo to deps.edn to be able to debug custom hooks from REPL

* Fix macroexpansion hook for with-temp* without values

* Test config (WIP)

* More misc fixes

* Disable :inline-def for tests

* More misc fixes

* Fix $ids and mbql-query kondo hooks.

* Fix with-temporary-setting-values with namespaced symbols

* More misc fixes

* Fix the rest of the easy ones

* Fix hook for mt/dataset

* Horrible hack to work around https://github.com/clj-kondo/clj-kondo/issues/1773 . Custom linter for mbql-query macro

* Fix places calling mbql-query with a keyword table name

* Fix the last few errors in test/

* Fix errors in enterprise/test and shared/test

* Fix driver test errors

* Enable linters on CI

* Enable unresolved-namespace linter for tests

* Appease the namespace linter again

* Test fixes

Addresses part of https://github.com/metabase/metabase/issues/8679

Pivot tables can have subqueries that run to create tallies. We do not
hold the entirety of resultsets in memory so we have a bit of an
inversion of control flow: connections are opened, queries run, and
result sets are transduced and then the connection closed.

The error here was that subsequent queries for the pivot were run while
the first connection is still held open. But the connection is no longer
needed. But enough pivots running at the same time in a dashboard can
create a deadlock where the subqueries need a new connection, but the
main queries cannot be released until the subqueries have completed.

Also, rf management is critical. It's completion arity must be called
once and only once. We also have middleware that need to be
composed (format, etc) and others that can only be composed
once (limit). We have to save the original reducing function before
composition (this is the one that can write to the download writer, etc)
but compose it each time we use it with `(rff metadata)` so we have the
format and other middleware. Keeping this distinction in mind will save
you lots of time. (The limit query will ignore all subsequent rows if
you just grab the output of `(rff metadata)` and not the rf returned
from the `:rff` key on the context.

But this takes the following connection management:

```
tap> "OPENING CONNECTION 0"
tap> "already open: "
  tap> "OPENING CONNECTION 1"
  tap> "already open: 0"
  tap> "CLOSING CONNECTION 1"
  tap> "OPENING CONNECTION 2"
  tap> "already open: 0"
  tap> "CLOSING CONNECTION 2"
  tap> "OPENING CONNECTION 3"
  tap> "already open: 0"
  tap> "CLOSING CONNECTION 3"
tap> "CLOSING CONNECTION 0"
```

and properly sequences it so that connection 0 is closed before opening
connection 1.

It hijacks the executef to just pass that function into the reducef part
so we can reduce multiple times and therefore control the
connections. Otherwise the reducef happens "inside" of the executef at
which point the connection is closed.

Care is taken to ensure that:
- the init is only called once (subsequent queries have the init of the
rf overridden to just return `init` (the acc passed in) rather than
`(rf)`
- the completion arity is only called once (use of `(completing rf)` and
the reducing function in the subsequent queries is just `([acc] acc)`
and does not call `(rf acc)`. Remember this is just on the lower
reducing function and all of the takes, formats, etc _above_ it will
have the completion arity called because we are using transduce. The
completion arity is what takes the volatile rows and row counts and
actually nests them in the `{:data {:rows []}` structure. Without
calling that once (and ONLY once) you end up with no actual
results. they are just in memory.

* dir locals for api/let-404

* Driver supports persisted model

* PersistedInfo model

far easier to develop this model with the following sql:

```sql
create table persisted_info (
   id serial primary key not null
  ,db_id int references metabase_database(id) not null
  ,card_id int references report_card(id) not null
  ,question_slug text not null
  ,query_hash text not null
  ,table_name text not null
  ,active bool not null
  ,state text not null
  ,UNIQUE (db_id, card_id)
)

```
and i'll make the migration later. Way easier to just dorp table, \i
persist.sql and keep developing without worrying about the migration
having changed so it can't rollback, SHAs, etc

* Persisting api (not making/deleting tables yet)

http POST "localhost:3000/api/card/4075/persist" Cookie:$COOKIE -pb
http DELETE "localhost:3000/api/card/4075/persist" Cookie:$COOKIE -pb

useful from commandline (this is httpie)

* Pull format-name into ddl.i

* Postgres ddl

* Hook up endpoints

* move schema-name into interface

* better jdbc connection management

* Hotswap peristed tables into qp

* clj-kondo fixes

* docstrings

* bad alias in test infra

* goodbye testing format-name function

left over. everything uses ddl.i/format-name and this rump was left

* keep columns in persisted info

columns that are in the persisted query. I thought about a tuple of
[col-name type] instead of just the col-name. I didn't do this this type
because I want to ensure that we compute the db-type in ONLY ONE WAY
ever and i wasn't ready to commit to that yet. I'm not sure this is
necessary in the future so it remains out now.

Context: we hot-swap the persisted table in for the original
query. Match up on query hash remaining the same. It continues to use
the metadata from the original query and just `select cols from table`

* Add migration for persisted_info table

also removes the db_id. Don't know why i was thinking that was
necessary. also means we don't need another unique constraint on (db_id,
card_id) since we can just mark the card_id as unique. no idea what i
was thinking.

* fix ns in a sad manner :(

far better to just have no alias to indicate it is required for side
effects.

* Dont hardcode a card-id :(:(:( my B

* copy the PersistedInfo

* ns cleanup, wrong alias, reflection warning

* Check that state of persisted_info is persisted

* api to enable persistence on a db

i'm not wild about POST /api/database/:id/persist and POST
/api/database/:id/unpersist but carrying on. left a note about it.

So now you can enable persistence on a db, enable persistence on a model
by posting to api/card/:id/persist and everything works.

What does not work yet is the unpersisting or re-persisting of models
when using the db toggle.

* Add refresh_begin and refresh_end to persisted_info

This information helps us with two bits:
- when we need to chunk refreshing models, this lets us order by
staleness so we can refresh a few models and pick up later
- if we desire, we can look at the previous elapsed time of refreshes
and try to gauge amount of work we want. This gives us a bit of
look-ahead. We can of course track our progress as we go but there's no
way to know if the next refresh might take an hour. This gives us a bit
of insight.

* Refresh tables every 8 hours ("0 0 0/8 * * ? *")

Tables are refreshed every 8 hours. There is one single job doing this
named "metabase.task.PersistenceRefresh.job" but it has 0 triggers by
default. Each database with persisted models will add a trigger to this
to refresh those models every 8 hours.

When you unpersist a model, it will immediately remove the table and
then delete the persisted_info record.

When you mark a database as persist false, it will immediately mark all
persisted_info rows as inactive and deleteable, and unschedule its
trigger. A background thread will then start removing the tables.

* Schedule refreshing on startup, watching for already scheduled

does not allow for schedule changes but that's a future endeavor

* appease our linter overlords

* Dynamic var to inbhit persistence when refreshing

also, it checked the state against "active" instead of "persisted" which
is really freaky. how has this worked in the past if thats the case?

* api docstrings on card persist

* docstring

* Don't sync the persisted schemas

* Fix bad sql when no deleteable rows

getting error with bad sql when there were no ids

* TaskHistory for refreshing

* Add created_at to persist_info table

helpful if this ever ends up in the audit section

* works on redshift

hooked up the hierarchy and redshift is close enought that it just works

* Remove persist_info record after deleting "deleteable"

* Better way to check that something exists

* POST /api/<card-id>/refresh

api to refresh a model's persisted record

* return a 204 from refreshing

* Add buttons to persist/unpersist a database and a model for PoC (#21344)

* Redshift and postgres report true for persist-models

there are separate notions of persistence is possible vs persistence is
enabled. Seems like we're just gonna check details for enabled and rely
on the driver multimethod for whether it is possible.

* feature for enabled, hydrate card with persisted

two features: :persist-models for which dbs support it, and
:persist-models-enabled for when that option is enabled.

POST to api/<card-id>/unpersist

hydrate persisted on cards so FE can display persist/unpersist for
models

* adjust migration number

* remove deferred-tru 



* conditionally hydrate persisted on models only

* Look in right spot for persist-models-enabled

* Move persist enabled into options not details

changing details recomposes the pool, which is especially bad now that
we have refresh tasks going on reusing the same connection

* outdated comment

* Clean up source queries from persisted models

their metadata might have had [:field 19 nil] field_refs and we should
substitute just [:field "the-name" {:base-type :type/Whatever-type}
since it will be a select from a native query.

Otherwise you get the following:

```
2022-03-31 15:52:11,579 INFO api.dataset :: Source query for this query is Card 4,088
2022-03-31 15:52:11,595 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,596 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,611 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4070 nil] for field "category.catid" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4068 nil] for field "category.catgroup" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,622 WARN middleware.fix-bad-references :: Bad :field clause [:field 4071 nil] for field "category.catname" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
2022-03-31 15:52:11,623 WARN middleware.fix-bad-references :: Bad :field clause [:field 4069 nil] for field "category.catdesc" at [:fields]: clause should have a :join-alias. Unable to infer an appropriate join. Query may not work as expected.
```
I think its complaining that that table is not joined in the query and
giving up.

While doing this i see we are hitting the database a lot:

```
2022-03-31 22:52:18,838 INFO api.dataset :: Source query for this query is Card 4,111
2022-03-31 22:52:18,887 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
2022-03-31 22:52:18,918 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
2022-03-31 22:52:18,930 INFO middleware.fetch-source-query :: Substituting cached query for card 4,111 from metabase_cache_1e483_229.model_4111_redshift_c
```

I tried to track down why we are doing this so much but couldn't get
there.

I think I need to ensure that we are using the query store annoyingly :(

* Handle native queries

didn't nest the vector in the `or` clause correctly. that was truthy
only when the mbql-query local was truthy. Can't put the vector `[false
mbql-query]` there and rely on that behavior

* handle datetimetz in persisting

* Errors saved into persisted_info

* Reorder migrations to put v43.00-047 before 048

* correct arity mismatch in tests

* comment in refresh task

* GET localhost:3000/api/persist

Returns persisting information:
- most information from the `persist_info` table. Excludes a few
columns (query_hash, question_slug, created_at)
- adds database name and card name
- adds next fire time from quartz scheduling

```shell
❯ http GET "localhost:3000/api/persist" Cookie:$COOKIE -pb
[
    {
        "active": false,
        "card_name": "hooking reviews to events",
        "columns": [
            "issue__number",
            "actor__login",
            "user__login",
            "submitted_at",
            "state"
        ],
        "database_id": 19,
        "database_name": "pg-testing",
        "error": "No method in multimethod 'field-base-type->sql-type' for dispatch value: [:postgres :type/DateTimeWithLocalTZ]",
        "id": 4,
        "next-fire-time": "2022-04-06T08:00:00.000Z",
        "refresh_begin": "2022-04-05T20:16:54.654283Z",
        "refresh_end": "2022-04-05T20:16:54.687377Z",
        "schema_name": "metabase_cache_1e483_19",
        "state": "error",
        "table_name": "model_4077_hooking_re"
    },
    {
        "active": true,
        "card_name": "redshift Categories",
        "columns": [
            "catid",
            "catgroup",
            "catname",
            "catdesc"
        ],
        "database_id": 229,
        "database_name": "redshift",
        "error": null,
        "id": 3,
        "next-fire-time": "2022-04-06T08:00:00.000Z",
        "refresh_begin": "2022-04-06T00:00:01.242505Z",
        "refresh_end": "2022-04-06T00:00:01.825512Z",
        "schema_name": "metabase_cache_1e483_229",
        "state": "persisted",
        "table_name": "model_4088_redshift_c"
    }
]

```

* include card_id in /api/persist

* drop table if exists

* Handle rescheduling refresh intervals

There is a single global value for the refresh interval. The API
requires it to be 1<=value<=23. There is no validation if someone
changes the value in the db or with an env variable. Setting this to a
nonsensical value could cause enormous load on the db so they shouldn't
do that.

On startup, unschedule all tasks and then reschedule them to make sure
that they have the latest value.

One thing to note: there is a single global value but i'm making a task
for each database. Seems like an obvious future enhancement so I don't
want to deal with migrations. Figure this gives us the current spec
behavior to have a trigger for each db with the same value and lets us
get more interesting using the `:options` on the database in the
future.

* Mark as admin not internal

lets it show up in `api/setting/` . I'm torn on how special this value
is. Is it the setting code's requirement to invoke the reschedule
refresh triggers or should that be on the setting itself.

It feels "special" and can do a lot of work from such just setting an
integer. There's a special endpoint to set it which is aware, and thus
would be a bit of an error to set this setting through the more
traditional setting endpoint

* Allow for "once a day" refresh interval

* Global setting to enable/disable

post api/persist/enable
post api/persist/disable

enable allows for other scheduling operations (enabling on a db, and
then on a model).

Disable will
- update each enabled database and disable in options
- update each persisted_info record and set it inactive and state
deleteable
- unschedule triggers to refresh
- schedule task to unpersist each model (deleting table and associated
pesisted_info row)

* offset and limits on persisted info list

```shell
http get "localhost:3000/api/persist?limit=1&offset=1" Cookie:$COOKIE -pb
{
    "data": [
        {
            "active": true,
            "card_id": 4114,
            "card_name": "Categories from redshift",
            "columns": [
                "catid",
                "catgroup",
                "catname",
                "catdesc"
            ],
            "database_id": 229,
            "database_name": "redshift",
            "error": null,
            "id": 12,
            "next-fire-time": "2022-04-08T00:00:00.000Z",
            "refresh_begin": "2022-04-07T22:12:49.209997Z",
            "refresh_end": "2022-04-07T22:12:49.720232Z",
            "schema_name": "metabase_cache_1e483_229",
            "state": "persisted",
            "table_name": "model_4114_categories"
        }
    ],
    "limit": 1,
    "offset": 1,
    "total": 2
}
```

* Include collection id, name, and authority level

* Include creator on persisted-info records

* Add settings to manage model persistence globally (#21546)

* Common machinery for running steps

* Add model cache refreshes monitoring page (#21551)

* don't do shenanigans

* Refresh persisted and error persisted_info rows

* Remarks on migration column

* Lint nits (sorted-ns and docstrings)

* Clean up unused function, docstring

* Use `onChanged` prop to call extra endpoints (#21593)

* Tests for persist-refresh

* Reorder requires

* Use quartz for individual refreshing for safety

switch to using one-off jobs to refresh individual tables. Required
adding some job context so we know which type to run.

Also, cleaned up the interface between ddl.interface and the
implementations. The common behaviors of advancing persisted-info state,
setting active, duration, etc are in a public `persist!` function which
then calls to the multimethod `persist!*` function for just the
individual action on the cached table.

Still more work to be done:
- do we want creating and deleting to be put into this type of system?
Quite possible
- we still don't know if a query is running against the cached table
that can prevent dropping the table. Perhaps using some delay to give
time for any running query to finish. I don't think we can easily solve
this in general because another instance in the cluster could be
querying against it and we don't have any quick pub/sub type of
information sharing. DB writes would be quite heavy.
- clean up the ddl.i/unpersist method in the same way we did refresh and
persist. Not quite clear what to do about errors, return values, etc.

* Update tests with more job-info in context

* Fix URL type conflicts

* Whoops get rid of our Thread/sleep test :)

* Some tests for the new job-data, clean up task history saving

* Fix database model persistence button states (#21636)

* Use plain database instance on form

* Fix DB model persistence toggle button state

* Add common `getSetting` selector

* Don't show caching button when turned off globally

* Fix text issue

* Move button from "Danger zone"

* Fix unit test

* Skip default setting update request for model persistence settings (#21669)

* Add a way to skip default setting update request

* Skip default setting update for persistence

* Add changes for front end persistence

- Order by refresh_begin descending
- Add endpoint /persist/:persisted-info-id for fetching a single entry.

* Move PersistInfo creation into interface function

* Hide model cache monitoring page when caching is turned off (#21729)

* Add persistence setting keys to `SettingName` type

* Conditionally hide "Tools" from admin navigation

* Conditionally hide caching Tools tab

* Add route guard for Tools

* Handle missing settings during init

* Add route for fetching persistence by card-id

* Wrangling persisted-info states

Make quartz jobs handle any changes to database.
Routes mark persisted-info state and potentially trigger jobs.
Job read persisted-info state.

Jobs

- Prune
-- deletes PersistedInfo `deleteable`
-- deletes cache table

- Refresh
-- ignores `deletable`
-- update PersistedInfo `refreshing`
-- drop/create/populate cache table

Routes

card/x/persist
- creates the PersistedInfo `creating`
- trigger individual refresh

card/x/unpersist
- marks the PersistedInfo `deletable`

database/x/unpersist
- marks the PersistedInfos `deletable`
- stops refresh job

database/x/persist
- starts refresh job

/persist/enable
- starts prune job

/persist/disable
- stops prune job
- stops refresh jobs
- trigger prune once

* Save the definition on persist info

This removes the columns and query_hash columns in favor of definition.

This means, that if the persisted understanding of the model is
different than the actual model during fetch source query we won't
substitute.

This makes sure we keep columns and datatypes in line.

* Remove columns from api call

* Add a cache section to model details sidebar (#21771)

* Extract `ModelCacheRefreshJob` type

* Add model cache section to sidebar

* Use `ModelCacheRefreshStatus` type name

* Add endpoint to fetch persistence info by model ID

* Use new endpoint at QB

* Use `CardId` from `metabase-types/api`

* Remove console.log

* Fix `getPersistedModelInfoByModelId` selector

* Use `t` instead of `jt`

* Provide seam for prune testing

- Fix spelling of deletable

* Include query hash on persisted_info

we thought we could get away with just checking the definition but that
is schema shaped. So if you changed a where clause we should invalidate
but the definition would be the same (same table name, columns with
types).

* Put random hash in PersistedInfo test defaults

* Fixing linters

* Use new endpoint for model cache refresh modal (#21742)

* Use new endpoint for cache status modal

* Update refresh timestamps on refresh

* Move migration to 44

* Dispatch on initialized driver

* Side effects get bangs!

* batch hydrate :persisted on cards

* bang on `models.persisted-info/make-ready!`

* Clean up a doc string

* Random fixes: docstrings, make private, etc

* Bangs on side effects

* Rename global setting to `persisted-models-enabled`

felt awkward (enabled-persisted-models) and renamed to make it a bit
more natural. If you are developing you need to set the new value to
true and then your state will stay the same

* Rename parameter for site-uuid-str for clarity

* Lint cleanups

interesting that the compojure one is needed for clj-kondo. But i guess
it makes sense since there is a raw `GET` in `defendpoint`.

* Docstring help

* Unify type :type/DateTimeWithTZ and :type/DateTimeWithLocalTZ

both are "TIMESTAMP WITH TIME ZONE". I had got an error and saw that the
type was timestamptz so i used that. They are synonyms although it might
require an extension.

* Make our old ns linter happy

Co-authored-by: Alexander Polyankin <alexander.polyankin@metabase.com>
Co-authored-by: Anton Kulyk <kuliks.anton@gmail.com>
Co-authored-by: Case Nelson <case@metabase.com>

This issue has a simple fix but a convoluted story. The new bigquery
driver handles multiple schemas and puts that schema (dataset-id) in the
normal spot on a table in our database. The old driver handled only a
single schema by having that dataset-id hardcoded in the database
details and leaving the schema slot nil on the table row.

```clojure
;; new driver describe database:
[{:name "table-1" :schema "a"}
 {:name "table-2" :schema "b"}]

;; old driver describe database (with dataset-id "a" on the db):
[{:name "table-1" :schema nil}]
```

So if you started on the new driver and then downgraded for some reason,
the table sync would see you had tables with schemas, but when it
enumerated the tables in the database on the next sync, would see tables
without schemas. It did not unify these two together, nor did it archive
the tables with a schema. You ended up with both copies in the
database, all active.

```clojure
[{:name "table-1" :schema "a"}
 {:name "table-2" :schema "b"}
 {:name "table-1" :schema nil}]
```

If you then tried to migrate back to the newer driver, we migrated them
as normal: since the old driver only dealt with one schema but left it
nil, put that dataset-id on all of the tables connected to this
connection.

But since the new driver and then the old driver created copies of the
same tables, you would end up with a constraint violation: tables with
the same name and, now after the migration, the same schema. Ignore this
error and the sync in more recent versions will correctly inactivate the
old tables with no schema.

```clojure
[{:name "table-1" :schema "a"}  <-|
 {:name "table-2" :schema "b"}    | constraint violation
 {:name "table-1" :schema "a"}] <-|

;; preferrable:
[{:name "table-1" :schema "a"}
 {:name "table-2" :schema "b"}
 {:name "table-1" :schema nil :active false}]
```

* Migrations for the new 'Annotate Useful Events' project

Users will be able to create 'events' in timelines, which are accessible via collections.

Two migrations are introduced to facilitate the new feature: timeline table, and event table. The following columns
are presumed necessary based on the design document:

** TimeLine
    - id
    - name
    - description
    - collection_id
    - archived
    - creator_id
    - created_at
    - updated_by ??
    - updated_at

*** Event
    - id
    - timeline_id
    - name
    - description markdown (max length 255 for some reason)
    - date
    - time is optional
    - timezone (optional)
    - icon (set of 6)
    - archived status
    - created by
    - created at
    - created through: api or UI
    - modified by ??
    - modified at

* Changes to events schema

- add icon onto timeline
- make icon on event nullable
- just move to a single timestamp on event, no boolean or anything
- rename event table to `timeline_events` since "event" is so generic

* dir-locals indentation

* Timeline api and model

- patched up the migration to make collection_id nullable since that is
the root collection
- followed the layout of api/metric and did the generic model stuff

* Select keys with keywords, not destructured nils :(

also, can't return the boolean from the update, but select it again

* Enable the automatic updated_at

* Boolean for "time matters" and string timezone column on events

* clean up migration, rename modified_at -> updated_at

for benefit of `:timestamped? true` bits

* basic timeline event model namespace

* Timeline Events initial API

Just beginning the basics for the API for timeline events.

 - need to find a way to check permissions
 - have to still check if the endpoint is returning stuff properly

* Singularize timeline_event tablename

* Timeline events api hooked up

* Singularize timeline event api namespace

* unused comment

* indent spec maps on routes

* Make name optional on timeline PUT

* Update collection api for timelines endpoints

- add /root/timelines endpoint for when collection id is null
- add /:id/timelines endpoint

Created a hydration function to hydrate the timeline events when fetching timelines.

* TimelineEvent permissions

- crate a permissions objects set via the timeline's permissions

* Move to using new permissions setup

previously was explicitly checking permissions of the underlying
timeline from the event, but now we just do the standard read/write
check on the event, and the permissions set knows it is based on the
permission set of the underlying timeline (which is based on the
permissions set of the underlying collection)

* Items api includes timelines

* Strip of icon from other rows

* Indices on timeline and timeline_event

timeline access by collection_id
timeline_event by both timeline_id and (timeline_id, timestamp) for when
looking for events within a certain range. We will always have been
narrowed by timeline ids (per the collection containing the question, or
by manually enabling certain timelines) so we don't need a huge index on
the timestamp itself.

* Skeleton of range-based query

* Initial timeline API tests

Began with some simple Auth tests and basic GET tests, using a Collection with an id as well as the special root collection.

* Fix docstring on api timeline namespace

* Put timeline's events at `:events` not `timeline-events`

* Add api/card/:id/timelines

At the moment api/card/:id/timelines and api/collection/:id/timelines do
the same thing: they return all of the timelines in the collection or in
the collection belonging to the card. In the future this may drift so
they share an implementation at the moment.

* Put creator on timeline

on api endpoints for timeline by id and timelines by card and collection

* Hydrate creator on events

Bit tricky stuff here. The FE wants the events at "events" not
"timeline-events" i guess due to the hyphen that js can never quite be
sure isn't subtraction.

To use the magic of nested hydration

`(hydrate timelines [:events :creator])`,

the events have to be put at the declared hydration key. I had wanted
the hydration `:timeline-events` for clarity in the code but want
`:events` to be the key. But when you do this, the hydration thinks it
didn't do any work and cannot add the creator. So the layout of the
datastructure is tied to the name of the hydration key. Makes sense but
I wish I had more control since `:events` is so generic.

* Add include param check to allow timeline GET to hydrate w/ events.

The basic check is in place for include=events in the following ways:

GET /api/timeline/:id?include=events
GET /api/collection/root/timelines?include=events

If include param is omitted, timelines are hydrated only with the creator. Events are always hydrated with creator.

* fix hyphen typo in api doc

* Change schema for include=events to s/enum to enforce proper param

Had used just a s/Str validation, which allows the include parameter value to be anything, which isn't great. So,
switched it to an enum. Can always change this later to be a set of valid enums, similar to the `model` parameter on
collection items.

* Fixed card/:id/timelines failing w/ wrong arity

The `include` parameter is passed to `timeline/timelines-for-collection` to control hydration of the events for that
timeline. Since the card API currently calls this function too, it requires that parameter be passed along.

As well, I abstracted the `include-events-schema` and have it in metabase.api.timeline. Not married to it, but since
the schema was being used across timeline, collection, and card API, it seems like a good idea to define it in one
place only. Not sure if it should be metabase.api.timeline or metabase.models.timeline

* used proper migration numbers (claimed in migrations slack channel)

* Add archived=true functionality

There's one subtle issue remaining, and that is that when archived=true, :creator is not hydrated on the events.

I'll look into fixing that.

In the meantime, the following changes are made:

 - :events returs empty list `[]` instead of `null` when there are no events in a timeline
 - archived events can be fetched via `/api/timeline/:id?include=events&archived=true`
 - similarly, archived events can be fetched with:
   - `/api/collection/root|:id/timelines?include=events&archived=true`
   - `/api/card/:id/timelines?include=events&archived=true`

Just note the caveat for the time being (no creator hydrated on events when fetching archived) Fix pending.

* Altered the hydration so creator is always part of a hydrated event

Adjusted the hydration of timelines as follows:

- hydration always grabs all events
- at the endpoint's implementation function, the timeline(s) are altered by filtering on archived true or false

Not sure if this is the best approach, but for now it should work

* Create GET endpoint for /api/timeline and allow archived=true

Added a missed GET endpoint for fetching all the timelines (optionally fetch archived)

* reformat def with docstring

* Timeline api updated to properly filter archived/un-archived

Use archived=true pattern in the query params for the timeline endpoint to allow FE to get what they need.

Work in progress on the API tests too.

* Timeline-events API tests

Timeline Events API endpoint tests, at least the very basics.

* Timeline Hydration w/ Events tests

Added a test for Events hydration in the timeline API test namespace.

* TimelineEvent Model test NS

May be a bit unnecessary to test the hydrate function, but the namespace is there in case we need to add more tests
over time.

Also adjusted out a comment and added a library to timeline_test ns

* Added metabase.models.timeline-test NS

Once again, this may be a bit unnecessary as a test ns for now, but could be the right place to add tests if the
feature grows in the future.

* Clean up handling of archived

- we only want to show archived events when viewing a timeline by id and
specifically asking for archived events. Presumably this is some
administrative screen. We don't want to allow all these options when
viewing a viz as its just overload and they are archived for a
reason. If FE really want them they can go by id of the timeline. Note
it will return all events, not just the archived ones.
- use namespaced keywords in the options map. Getting timelines
necessarily requires getting their events sometimes. And to avoid
confusion about the archived state, we have options like
`:timeline/events?`, `:timeline/archived?`, `:events/start`,
`:events/end`, and `:events/all?` so they are all in one map and not
nested or ambiguous.

* Clean up some tests

noticable differences: timelines with archived=true return all events as
if it is "include archived" rather than only show archived.

* PUT /api/timeline/:id now archives all events when TL is archived

We can archive/un-archive timelines, and all timeline events associated with that timeline will follow suit.

This does lose state when, for example, there is a mix of archived/un-archived events, as there is no notion of
'archived-by' to check against. But, per discussion in the pod-discovery slack channel, this is ok for now. It is also
how we handle other archiving scenarios anyway, with items on collections being archived when a collection is archived.

* cleanup tests

* Include Timeline and TimelineEvent in models

* Add tt/WithTempDefaults impls for Timeline/TimelineEvent

lets us remove lots of the `(merge defaults {...})` that was in the code.

Defaults are

```clojure
   Timeline
   (fn [_]
     {:name       "Timeline of bird squawks"
      :creator_id (rasta-id)})

   TimelineEvent
   (fn [_]
     {:name         "default timeline event"
      :timestamp    (t/zoned-date-time)
      :timezone     "US/Pacific"
      :time_matters true
      :creator_id   (rasta-id)})
```

* Add timeline and timeline event to copy infra

* Timeline Test checking that archiving a Timeline archives its events

A Timeline can be archived and its events should also be archived.
A Timeline can also be unarchived, which should also unarchive the events.

* Docstring clarity on apis

* Remove commented out form

* Reorder migrations

* Clean ns

clj-kondo ignores those but our linter does not

* Correct casing on api schema for include

* Ensure cleanup of timeline from test

* DELETE for timeline and timeline-event

* Poison collection items timeline query when is_pinned

timelines have no notion of pinning and were coming back in both queries
for pinned and not pinned. This adds a poison clause 1=2 when searching
for pinned timelines since none are but they don't have a column saying
that.

* Clean up old comment and useless tests

comment still said to poison the query when getting pinned state and
that function had been added.

Tests were asserting count = 2 and also the set of names had two things
in them. Close enough since there are no duplicate names

* Use TemporalString schema on timeline event routes

Co-authored-by: dan sutton <dan@dpsutton.com>

* Handle nested queries which have agg at top level and nested

Previously when matching columns in the outer query with columns in the
inner query we had use id, and then recently, field_ref. This is
problematic for aggregations.

Consider https://github.com/metabase/metabase/issues/19403

The mbql for this query is

```clojure
{:type :query
 :query {:aggregation [[:aggregation-options
                        [:count]
                        {:name "count"}]
                       [:aggregation-options
                        [:avg
                         [:field
                          "sum"
                          {:base-type :type/Float}]]
                        {:name "avg"}]]
         :limit 10
         :source-card-id 1960
         :source-query {:source-table 1
                        :aggregation [[:aggregation-options
                                       [:sum
                                        [:field 23 nil]]
                                       {:name "sum"}]
                                      [:aggregation-options
                                       [:max
                                        [:field 28 nil]]
                                       {:name "max"}]]
                        :breakout [[:field 26 nil]]
                        :order-by [[:asc
                                    [:field 26 nil]]]}}
 :database 1}
```

The aggregations in the top level select will be type checked as :name
"count" :field_ref [:aggregation 0]. The aggregations in the nested
query will be turned into :name "sum" :field_ref [:aggregation 0]! This
is because aggregations are numbered "on their level" and not
globally. So when the fields on the top level look at the metadata for
the nested query and merge it, it unifies the two [:aggregation 0]
fields but this is INCORRECT. These aggregations are not the same, they
just happen to be the first aggregations at each level.

Its illustrative to see what a (select * from (query with aggregations))
looks like:

```clojure
{:database 1
 :query {:source-card-id 1960
         :source-metadata [{:description "The type of product, valid values include: Doohicky, Gadget, Gizmo and Widget"
                            :semantic_type :type/Category
                            :coercion_strategy nil
                            :name "CATEGORY"
                            :field_ref [:field 26 nil]
                            :effective_type :type/Text
                            :id 26
                            :display_name "Category"
                            :fingerprint {:global {:distinct-count 4
                                                   :nil% 0}
                                          :type {:type/Text {:percent-json 0
                                                             :percent-url 0
                                                             :percent-email 0
                                                             :percent-state 0
                                                             :average-length 6.375}}}
                            :base_type :type/Text}
                           {:name "sum"
                            :display_name "Sum of Price"
                            :base_type :type/Float
                            :effective_type :type/Float
                            :semantic_type nil
                            :field_ref [:aggregation 0]}
                           {:name "max"
                            :display_name "Max of Rating"
                            :base_type :type/Float
                            :effective_type :type/Float
                            :semantic_type :type/Score
                            :field_ref [:aggregation 1]}]
         :fields ([:field 26 nil]
                  [:field
                   "sum"
                   {:base-type :type/Float}]
                  [:field
                   "max"
                   {:base-type :type/Float}])
         :source-query {:source-table 1
                        :aggregation [[:aggregation-options
                                       [:sum
                                        [:field 23 nil]]
                                       {:name "sum"}]
                                      [:aggregation-options
                                       [:max
                                        [:field 28 nil]]
                                       {:name "max"}]]
                        :breakout [[:field 26 nil]]
                        :order-by [[:asc
                                    [:field 26 nil]]]}}
 :type :query
 :middleware {:js-int-to-string? true
              :add-default-userland-constraints? true}
 :info {:executed-by 1
        :context :ad-hoc
        :card-id 1960
        :nested? true
        :query-hash #object["[B" 0x10227bf4 "[B@10227bf4"]}
 :constraints {:max-results 10000
               :max-results-bare-rows 2000}}
```

The important bits are that it understands the nested query's metadata
to be

```clojure
{:name "sum"
 :display_name "Sum of Price"
 :field_ref [:aggregation 0]}
{:name "max"
 :display_name "Max of Rating"
 :field_ref [:aggregation 1]}
```

And the fields on the outer query to be:
```clojure
([:field
  "sum"
  {:base-type :type/Float}]
 [:field
  "max"
  {:base-type :type/Float}])
```

So there's the mismatch: the field_ref on the outside is [:field "sum"]
but on the inside the field_ref is [:aggregation 0]. So the best way to
match up when "looking down" into sub queries is by id and then by name.

* Some drivers return 4.0 instead of 4 so make them all ints

* Handle dataset metadata in a special way

rather than trying to set confusing merge rules, just special case
metadata from datasets.

Previously, was trying to merge the "preserved keys" on top of the
normal merge order. This caused lots of issues. They were
trivial. Native display names are very close to the column name, whereas
mbql names go through some humanization. So when you select price
from (select PRICE ...) its an mbql with a nested native query. The
merge order meant that the display name went from "Price"
previously (the mbql nice name for the outer select) to "PRICE", the
underlying native column name. Now we don't special case the display
name (and other fields) of regular source-metadata.

Also, there were issues with mbql on top of an mbql dataset. Since it is
all mbql, everything is pulled from the database. So if there were
overrides in the nested mbql dataset, like description, display name,
etc, the outer field select already had the display name, etc. from the
database rather than allowing the edits to override from the nested
query.

Also, using a long `:source-query/dataset?` keyword so it is far easier
to find where this is set. With things called just `:dataset` it can be
quite hard to find where these keys are used. When using the namespaced
keyword, greping and finding usages is trivial. And the namespace gives
more context

* Fix logging utils

* Test fixes 

* Add new Settings (placeholders) for the new 0.41.0+ premium features

* Code cleanup

* Add new features to the `:premium-features` Setting (for FE consumption)

* Fix busted ns declarations

* Move sync executor to bespoke namespace

* Refingerprint fields on type change

* Check if can connect in refingerprint-field!

* docstring update

* Cleanup tests a bit

* Error handling

* Table field values endpoint on sync.concurrent executor

* ns sort api field

* ns cleanup

* Fix JUnit output not correctly stripping ANSI color codes

* Remove find-tests profiling log message

* 1-arg arity of metabase.test-runner/run

* Fix bug with metabase.test-runner/find-tests with a single test name

* Move *ignore-cached-results* to middleware

Had tried to adjust the binding to work, but the binding ended before
all of the bound functions and such were submitted to work, so when
that code actually executed the binding was no longer in
place. Annoying and it should be in middleware anyways

* Remove now unused import (cache became keyword in middleware not binding)

* Add type on collection

* Search with collection type

* Cleanup bounded heap accumulator

* Make search conditionally bump official collection items

* collection api and tests

* Put collection type onto hydrated cards on dashboards

* Move official collection type scoring into ee codebase

* ensure ee and oss agree when not official collection

* Mark Collections tree with type

* Remove unnecessary `and`s when no check for enhancements

* Tests for setting collection tree type

* Include hydration for collection type on dashboards

* Make sure created test collections are cleaned up

* Cleanup tests, don't search on collection_type

looks for all text fields and searches the term against that. official
would bring back all official types. no bueno

* Docstring on protocol and don't shadow comparator

* update to new ee impl var passing style

* Collection model ensures no type change on personal collection

* Check for personal collection when updating type

model checks for personal collection and rejects in the update but
doesn't check for children of personal collection.

* Update checking for unchangeable properties of a personal collection

* Cleanup: type collection tree, batched hydration, combine error checks

* Cleanup

* move bounded-heap accumulator to utils

* switch to test.check testing

* Bad test just to see what our CI will report

* remove purposeful failing test (was for CI)

* collection.type -> collection.authority_level

* Test the actual ee scoring impl, not the wrapped one

locally i had enhancements enabled so the tests were passing, but in
CI it was failing as it did not have enhancements enabled

* Port legacy data type migrations -> Liquibase

* Fix migration IDs

* Field type validation/error handling

* Have semantic type fallback to nil

* Fix semantic-type-migrations-test

* Fix migrations

* Revert accidental changes

* Semantic/* & Relation/* ancestor types

* Fix stray Relation/PK and Relation/FKs

* Semantic/* and Relation/* ancestor types

* cljs test fix

* Fix :require

* FE test fixes 

* Test fixes 

* prettier

* PR  f e e d b a c k

* Use medium size CircleCI image for Presto to prevent all the OOMs

* Backport dir-locals tweaks from hierarchy PR

* Redshift: only sync the test schema (faster CI and fix failure)

* Better error handling for sync in tests

* Revert accidental commit

* Redshift test fixes 

* Remove type/UNIX* and type/ISO8601* from frontend

* Set effective-type and coercion strategy when syncing

these values will most likely only be available when using tests, as
metadata is very unlikely to have this. As far as I can tell the
toucannery apparatus is the only bit that has this. Its quite
artificial. I'm not sure if this is a good idea.

* :type/UNIXTimestampSeconds and type/ISO8601DateTimeString out of type

remove the coercions from the type hierarchy. This brought up a
strange issue that has been present for a while: all liquidbase
migrations run and then all data migrations run. They are not
interleaved. This allows for the following scenario:

- data migration migrates all X values to Y in version 26
- liquibase migration migrates all Y values to Z in version 28

But these are not run in version order, but all liquibase migrations
are run and then all data migrations are run. If you were on an old
version for a while, you could end up with Y values even though the
liquibase migration which comes "after" the data migration turns all Y
values into Z values.

This impacts this change because a data migration in this way:
- liquibase changesets 287, 288, 289, and 290 remove all 'type/UNIX*'
and 'type/ISO8601*' semantic types. These were in 0.39
- data migration `migrate-field-types` added in 0.20.0 was looking for
special_type's of "timestamp_milliseconds" and migrating them to
"type/UNIXTimestampMilliseconds".

Since the liquibase runs before the migrate-field-types, it's possible
for a 'type/UNIX*' semantic type to reappear. And since these were
removed from the type system it would fail validation and blow up. In
this case it actually can't happen since the field its attempting to
migrate (special_type) no longer exists. But since the migrations are
not interleaved this problem still exists.

* whatever prettier

* Appease the machines

* Restore the unix and iso semantic types to hierarchy

migration tests still use these
`(impl/test-migrations [283 296] [migrate!] ...)`

A few things in tension: the field requires valid semantic_types,
these _were_ valid semantic_types until v39.

* Remove old "coercion" semantic types

* Migrate the v20 version semantic types for unix epoch

* Time travelling migrations

these target v20 and v29 which at the time they existed had a column
special type not semantic type. But these run after all of the schema
migrations, not interleaved, so they will have a semantic_type not
special_type when they actually run even though they target v20 or v29
data. strange world

* add migration's new checksum

* Backend feature flag for new field filters

* Feature flag new parameter options

When the "field-filter-operators-enabled?" flag is disabled we do the following:
1. Replace new operator options with old category and location/city, etc., options
   in the PARAMETER_OPTIONS list found in metabase/meta/Parameter.js
2. Hide numbers section in the PARAMETER_SECTIONS list found in
   metabase/meta/Dashboard.js
3. Return args as-is in the mapUIParameterToQueryParameter function
   found in metabase/meta/Parameter.js

React/UI code handles both old options and new options so doesn't need
to change. Old parameter types like "category" and "location/city" are
treated like "string/=" in the UI but retain their own parameter type
when used to send a new query.

* Fix FE issues caused by meta/Parameter refactor

* mock the field operator param flag to make tests pass

* add/fix cypress tests

* fix import in ParametersPopover

* update widget tag type

* Enable field filter operators for cypress tests

* Question marks are questionable

* Conditionally use category or string/= if field filters are enabled

* rmv mocks where we don't need them

* rmv mock from chained-filters test

* env vars as string in project.clj, alignment

Co-authored-by: dan sutton <dan@dpsutton.com>

* Set lein as the preferred build tool for Cider

* Add search-typeahead-enabled setting