* Add metrics and segments to entity store type

* Add "sensitive" field visibility type

* Add missing properties to MLv1 metadata classes

* Handle metrics and segments in `createEntitiesState`

* Convert metadata selector file to TypeScript

* Fix type errors

* Remove not used selectors

* Don't export private selectors

* Remove `getDatabases` selector

* Remove `getFields` selectors

* Fix types

* Use normalized entity types in metadata selectors

* enforce import order

* reorder all imports

* more import updates

* refactor data-permissions selectors

* pre refactor

* fix all typescript errors

* fixes

* fixes

* fixes

* use correct types

* RIP checksum

Now that users can edit the metadata the checksum prevents that and
would throw it away when edited. We'll treat the metadata as valid if it
conforms. The UI isn't really expected to modify type information anyways.

* Ns checker is a harsh disciplinarian

* Extract `fields` variable in dataset editor

* Allow extra onChange callback to form fields

* Only show metadata sidebar when field is selected

* Add onChange callback to semantic type picker

* Add action for overwriting results metadata

* Connect essential fields to form

* display_name
* description
* semantic_type

TBD: mapped database column for native datasets

* Fix missing import

* Fix sidebar inputs loosing focus while typing

* Ensure focused field is in sync

* Refactor FK and currency to be form fields

* Use `field.settings` to keep formatting changes

* Remove "display_as" from form definition

* Connect remaining fields

* Remove CurrencyPicker in favor of formatting input

* Fix incomplete field metadata

Results metadata doesn't contain everything we need for the editor.

'Global' metadata is the most complete, though it shouldn't be muted while in the editor, only once a user saves the changes

Now we only keep focused field ref in state and derive the metadata by merging global metadata and query's results metadata. Query results metadata is mutated by the editor, so merging them both will do the job

* Fix ViewSidebar prop type

* Fix prop type warnings

* Add default values for some fields

* Connect MappedFieldPicker

* Fix MappedFieldPicker closes after loading fields

Sidebar dependant on IDFields was rerendering once additional metadata was loaded, causing DataSelector to unmount

Fixed by moving IDFields subscription to FKTargetPicker directly (the only component that actually needs them)

* Hide metadata popovers in dataset editor

* Disable saving invalid dataset editor changes

Disables "Save changes" button if:
* there is at least one field with empty display name
* the underlying query is empty

* Add `onChangeSetting` prop to ColumnSettings

The existing `onChange` passes the whole settings object as an argument. On the other hand, `onChangeSetting` passes only the changed setting which is useful for storing metadata diff

* Keep field metadata diff in redux

* Ignore column width reset events in metadata editor

* Preserve metadata diff between query edits

* A few important changes:

- ensure that we don't throw away metadata and return an empty channel
in the card api if the query hasn't changed
- ensure api/dataset uses `:metadata/dataset-metadata`
- hydrate fields on api/table/card__id/query_metadata so that it has
target and fk information
- preserve id so this information can be recovered†

†: When a user enters an id we preserve it, but it also gets confusing
if that's the actual field's id or the user-identified id. This is
important when we drill down (and probably lots of other places too if
they start assuming an id means they know exactly what is going on).

A situation based on the orders table:

Suppose you make a native question

```sql
SELECT "PUBLIC"."ORDERS"."ID" AS "ID",
  "PUBLIC"."ORDERS"."USER_ID" AS "user_id_bro",  -- note this is renamed
  "PUBLIC"."ORDERS"."PRODUCT_ID" AS "PRODUCT_ID",
  "PUBLIC"."ORDERS"."SUBTOTAL" AS "SUBTOTAL",
  "PUBLIC"."ORDERS"."TAX" AS "TAX",
  "PUBLIC"."ORDERS"."TOTAL" AS "TOTAL",
  "PUBLIC"."ORDERS"."DISCOUNT" AS "DISCOUNT",
  "PUBLIC"."ORDERS"."CREATED_AT" AS "CREATED_AT",
"PUBLIC"."ORDERS"."QUANTITY" AS "QUANTITY"
FROM "PUBLIC"."ORDERS"
LIMIT 15
```

In my db, the "USER_ID" column is field 3 and points at field 22. Our db
knows that field 3 is called "USER_ID". But in the source query that
column is actually called "user_id_bro". And lots of places assume that
if there's a field's `:id` it knows what the name is. This is obviously
incorrect here. It is a very important distinction that this column "is
basically field 3 USER_ID" and this is literally the field "USER_ID".

So if you do a drilldown based on this information, the FE will issue a
query to "select field 3 from card__127" which expands to roughly
"select USER_ID from (select ... USER_ID as user_id_bro ...)" and this
query obviously will fail.

* Don't put nil on a channel, clean ns

i honestly don't know why the results-metadata and encrypt deps were
there. I didn't add them but they just got caught in the linter for some
reason. I don't understand it.

* Update tests to include target and has_field_values

* Correct final test

* Select enough information for FK to work in UI

* Pass mapped column's properties to BE

* Qualify which semantic_type FK/PK we remove

Previously always removed these semantic types. Now only remove when
they do not have an id there. One thing I'm confused about though is
that native fields should never have a semantic type anyways. There's no
way for one to end up there, certainly not FK or PK right? So i want to
delete this entire method because as I understand it it's not possible
to trigger its conditions except for the new user-entered information,
in which case we want to preserve it. But I have no idea why
Mr. Chesterton built this function so I don't want to delete it.

Bug reports and connecting them back to this change is so hard. Because
it can be things like "native query with an aggregation named "blah"
bucketed by month break when you drilldown into this column" and that's
because it includes a semantic_type that breaks something. It's all very
at-a-distance so hard to trace back to this.

* Fix failed queries handling

* Support custom column metadata editing

* Hide empty sections in semantic type picker

* Remove redundant import

* Fetch database ID fields in FKTargetPicker

* Allow aliasing EntityObjectLoader's entity prop

* Reset DataSelector's state when ID props change

* Fix MappedFieldPicker

* Fixed saved question picker regression

* Select description and settings when merging result_metadata

Annotate columns allows for settings to be preserved from
datasets. Example

```clojure
;; shows a little bar in the UI. quite nice
:settings {:show_mini_bar true, :decimals 0},
```

These settings were correctly preserved in annotate.clj but dropped in
the final merging in result_metadata.clj. Fun times in metadata land

* Let dataset revisions preserve metadata

* Only wrap form fields with custom onChange handler

This prevents form fields from extra re-renders

* Fix summarize sidebar behavior

* Fix e2e test

* Update cypress expectations

Previously expected really generic placeholder text in search boxes. But
in `api/table/card__id/query_metadata` we hydrate native columns with
`:has_field_values` so that when a user identifies an underlying field
on datasets, we have extra information.

This change comes about because the hydrate for `:has_field_values`
doesn't hit the db and look for field values but uses a heuristic.  And
this heuristic works for native fields and finds that they are
searchable because when they are text or textlike and improves the
search bar.

Code from models/field.clj below

```clojure
(defn- is-searchable?
  "Is this `field` a Field that you should be presented with a search widget for (to search its values)? If so, we can
  give it a `has_field_values` value of `search`."
  [{base-type :base_type}]
  ;; For the time being we will consider something to be "searchable" if it's a text Field since the `starts-with`
  ;; filter that powers the search queries (see `metabase.api.field/search-values`) doesn't work on anything else
  (or (isa? base-type :type/Text)
      (isa? base-type :type/TextLike)))

(defn- infer-has-field-values
  "Determine the value of `has_field_values` we should return for a `Field` As of 0.29.1 this doesn't require any DB
  calls! :D"
  [{has-field-values :has_field_values, :as field}]
  (or
   ;; if `has_field_values` is set in the DB, use that value; but if it's `auto-list`, return the value as `list` to
   ;; avoid confusing FE code, which can remain blissfully unaware that `auto-list` is a thing
   (when has-field-values
     (if (= (keyword has-field-values) :auto-list)
       :list
       has-field-values))
   ;; otherwise if it does not have value set in DB we will infer it
   (if (is-searchable? field)
     :search
     :none)))

(defn with-has-field-values
  "Infer what the value of the `has_field_values` should be for Fields where it's not set. See documentation for
  `has-field-values-options` above for a more detailed explanation of what these values mean."
  {:batched-hydrate :has_field_values}
  [fields]
  (for [field fields]
    (when field
      (assoc field :has_field_values (infer-has-field-values field)))))
```

* Ignore `has_field_values` for field literals

Field literals are field using a column title instead of a numeric ID. They're usually coming from native or nested queries.

Metadata propagation changes introduced for data models, made the BE send a guessed `has_field_values` property for field literals too. But searching through possible field values isn't properly supported yet and it's difficult to prevent `has_field_values` being sent to the FE, so this makes the FE ignore it for field literals

* Fix E2E test

* Fix field literal check

* Ensure `:has_field_values` only goes on fields with ids

when a user selects an id to back a native field, we want to hydrate the
field_value information and the target info. But _ONLY_ when they have
ids. The hydration for `:has_field_values` works on a heuristic, not by
hitting the db and seeing if we have field values and will give
information about native fields that do not have backing field information.

* Ensure no annotations on numeric field ids

we put the vector id on each field when there's no numeric id field. I
have no idea why and the comment `;; TODO -- what????` explains the
thinking there. So only annotate where we have numeric ids (fields that
we know how to identify fields in the db) and not when we have a made up
vector id

* Revert changes to FieldValuesWidget

* Fix character case

* Show `has_field_values` selector when ID is known

* Undo the testing for indiscriminant hydration

previously had put target and has_field_values on every native field
which doesn't work well for the UI. Had to clean up the remaining tests
to no longer expect it

* Select `:dataset` when creating a new question

when duplicating a question, we send over this info and need to select
it from the existing question.

* Ensure card is saved with boolean dataset info

largely just for testing which does not send a `:dataset false` property
in some cases. In regular usage there should always be an initial value

* Remove dead code

* Replace TODO with a meaningful comment

* Type Field type

* Select only what we need on the source card

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

* add PermissionSelect component

* add PermissionSidebar component

* add PermissionTable component

* add PermissionEditor component

* add useLeaveConfirmation hook

* add selectors for the collections permissions page

* add selectors for the data permissions pages

* add confirmation creators to use in selectors

* sandboxing permissions modal: redirect to parent route when closed

* Add PermissionsPageLayout component

* Adjust CollectionPermissionsModal to use new permissions UI

* Add CollectionPermissionsPage

* Add data permissions pages

* Adjust collection permissions modal route

* Adjust permissions routes according new pages structure

* update permissions actions, reducers

* adjust sandboxing

* remove old permissions code

* support admin colorScheme in Radio

* support admin colorScheme in TextInput

* support admin colorScheme in Tree

* update permissions specs

* fix data permissions url typo

* fix sandboxing

* add search debounce to permission pages

* exclude schemas from search, make it flat

* fix permissions for databases without schemas

* fix copy for native queries permissions, show table display_name instead of raw name

* fix before leave warning shown when change collections

* permissions sidebar: make database name clickable

* fix string groupIds lead to bugs in permissions editor

* fix disabling table permissions shows all tables permission disabled

* fix permissions copy

* fix permissions disabled tooltips copy

* fix, add specs

* fix permissions edit bar button colors, fix breadcrumbs on schemaless dbs

* fix official collection color, make sidebar tree scrollable

* fix scroll in permissions modal

* fix breadcrumbs

* change table icon

* create helpers for entity ids and permissions url creation

* remove hardcoded colors

* make confirmations an array

* store confirmAction in state

* add ModalRoute specs

* fix propTypes, typo

* build permission pages urls with helpers

* revert switching from ref to state for confirm action function due to react invokes it on a render

* move mapStateToProps to the top of the files

* add visual specs

* add more specs

* fix edit sandbox access action

* add cypress spec for sandboxing

* yarn remove flow-bin eslint-plugin-flowtype

* CI: remove fe-linter-flow

* Remove Flow from ESLint config

* Remove Flow type annotations and directives

* Simplify Flow types for React elements and components

* Import missing Flow types

* Remove lint-flow from package.json