* New CI image
* Upgrade to new Clojure CLI version + multi arch images + add git to Builder images [ci nocache]

* Enable saving to subcollection when access to parent is revoked (#14052)

* Formatting

* fix adding questions to dashboards when parent is not shared

* add lodash to dependencies

Co-authored-by: Chris Wu <chris@faros.ai>

* migrate to TypeScript, add type annotations
* include props definition
* ensure a11y
* support Storybook
* add a basic unit test

* Add support for parsing number parameters as numbers

* Kill the regex

* Add TextWidget array value support

* Fix some tests

* destructure

* build resources

* update translations for x.43.1

* migrate to TypeScript, add type annotations
* include props and state definition
* support Storybook
* add a basic unit test

* Update persist model api permissions for EE

On OSS, only Admins can enable model cache in `Settings/Cache`,
enable database model cache in `Settings/Databases` and cache individual models.

On EE/Pro, users with `Settings access` application permissions can enable model
cache in `Settings/Cache` and users with `Manage database` can enable model
cache for a database.

* Add tests for application permissions

Addressing
Running clj -X:deps tree in bin/build-drivers the cli version is getting picked up
as I switched clojure-cli versions between 1.11 and 1.10.

I think this is because including metabase/metabase-core as a local/root no longer
places the clojure dep within it at the top level.

Now, the docs say
> It is a top dep (top dep versions always win) or it is a new lib or a newer version of a known lib (Maven keeps only the first found version regardless of new/old)

And so I would expect to see org.clojure/clojure in the deps tree under
metabase/metabase-core but it's missing, which to me means clojure maybe treated
specially if the root deps don't specify it.

UUIDs are treated as strings and when checking is-empty/not-empty, we compare
them against the empty string.
But an empty string is not a valid UUID, so this comparison is useless for
Postgres where a UUID field cannot have such a value. Assuming that a UUID
field is only ever compared to the empty string in the context of
is-empty/non-empty checks, replacing it with nil is OK.

* Record startup times

updated `dev/start!`. `mb.core/init!` itself sets up the db, loads
plugins, and sets the init status so those were superfluous.

* clean up dev namespace

* testing doesn't call init so set a value

* SQL JDBC Delete Action [ci skip]

* Add note about parsing values [ci skip]

* transition to using mbql for actions/row/:action

- TODO: how to setup a qp store?
- TODO: better way to keywordize filters?

* some cleanup [ci skip]

* better message on delet statement conflicts

- and a test

* mt/with-everything-store -> seed store manually [ci skip]

* fix some test input

* get more tests passing

* add more validation tests

* privatize catch-throw

* plug in and alter tests to use with-actions-test-data

Co-authored-by: Bryan Maass <bryan.maass@gmail.com>

* Don't reschedule syncs on startup

Previously when we started up, for each database we compare its current
schedule to its schedule in Quartz, and update the triggers if they were
different. But we handle this change on db inserts and db updates so
this is completely unnecessary. We did this back when quartz was in
memory only and it remained when we persisted.

We can remove this complexity. On an app-db with 400 postgres databases,
this takes 46 seconds in the old world and appears as just another log
statement now.

Reports from the wild are even more distressing: the quartz tables got
full of dead tuples causing queries to the triggers to take a long
time. They also had 500+ databases. We query all triggers and linear
search for each database and this process exploded up to as much as 8
hours. Now we are not doing all of this extra work and startup should be
quick again.

* clj-kondo ignore unused private var

* appease ns linter

* remove clojure.core/binding and unused bindings

* Rename `default-schedules` -> `default-randomized-schedule`

makes it a bit more clear what's going on

* Put randomizer in transducer over reducible query; tests

basically a `run!` over a reducible query that also does a count for
logging. Tests now check the three options:
- default schedules, updated
- non-default schedules, not updated timestamp verified that it is
identical.
- default schedules but user controls schedule: not updated, timestamp
verified that it is identical.

* Prevent randomized schedules from returning a default value

it actually doesn't matter in the application but the tests are
asserting that the newly randomized schedule is not one of the defaults
it was changed from and that can fail sometimes. Doesn't matter
logically but no biggie to simplify the tests.

Otherwise they fail in (1/60)+(1/24) test runs which is actually
decently frequently with 26 test suites per commit

* Add stub `WritebackForm`

* Connect QB to writeback flag state

* Add `WritebackForm` to `QueryModals`

* Add "New row" button

* Add missing `Table` and `Field` type properties

* Generate simple form out of field metadata

* Fix form layout in a modal

* Add field fingerprint types

* Show "New row" button on raw tables too

* Allow overwriting `FieldValuesWidget` renderers

* FieldValuesWidget: prop to always use `TokenField`

* Add class names to `TokenField`

* Highlight input border on hover

* Add category field picker for writeback form

* Support initial values for `WritebackForm`

* Add editing mode to `ObjectDetail`

* Fix type errors

* Fix `CategoryFieldInput` `onChange` return value

* Ooops

Open redirects means doing some sso with a built-in redirect, and redirecting into an unhappy place (aka, a non-MB place) afterwards so that someone gets phished or other bad things happen. This is already prevented for OSS sso's but not EE - prevents this for EE sso's by forcing redirects to be in MB `site-url` set domain.

* Destructible version of `test-data` dataset for testing the actions stuff

* Revert new precondition

* Enable detail view for any row

* update url handling for row indexes

* fix types and tests

* add unit tests

* test updates

* more test fixes

* Fix backticks

* Fix color usage in ObjectDetail

* Use styled component

* Fix missing translation

* Fix accidental thing

Co-authored-by: Anton Kulyk <kuliks.anton@gmail.com>

* Extend types

* Refactor FormField

* Fix prop values

* Fix types

* Fix field type check

* add logic to limit which parameters are linkable

* Filter linked filters list to all field filter parameters

* use some destructuring

* Improve DateMonthYearWidget component

* migrate to TypeScript, add type annotations
* include props definition
* support Storybook
* add a basic unit test

It's too slow, it's O(n), make n go down a fair bit while we do a bigger fix and set up regular profiling just for this

* do not show restore sample database button to non-admins

* fix db list update after restoring the sample db

* Refactor CustomForm

* Extract common types to `metabase-types`

* Refactor StandardForm

* Fix form field rerenders

* Fix types

* Clean up null logic

* Update fn name

* Simplify convertParametersToFilters logic

* Rename method

* Turn `CollectionCreate` into functional component

* Convert `CollectionCreate` to TypeScript

* Add `onChangeField` to Form container

* Remove `redux-form` from `CollectionCreate`

* Fix collection form selector

* Move `CollectionEdit` to its own directory

* Move `CollectionEditForm` to its own file

* Turn `CollectionEdit` into functional component

* Migrate `CollectionEdit` components to TypeScript

* Fix `CollectionCreate`

* Fix `CollectionEdit`

* Fix optional `Form` `onChange` prop

* Move `DatabaseHelp` to `DatabaseStep`

* Remove `redux-form` usage from setup

* Fix unit tests

* Move engine state to redux

* Revert test changes

* Code style fix

* Start task scheduler at end of startup.

The sync db task does a lot of db work to be addressed in a later
commit. But it also begins firing tasks _during startup_. Lets prevent
syncs and other tasks while we are starting up until we have finished.

Scheduler can be set to standby mode while we wait on finishing up
initializing and then start after.

* docstring cleanup and comments

We are not running tests which exercise those *.stories files, hence it
is not fair to include them in the coverage analysis.