* Properly connect actions to search entity

* Add e2e test

* Make nil equality comparable
* Add Malli schema for filter expressions

* Fix dashboard filter values coercsion for actions

* update custom action e2e test to cover parameters

---------

Co-authored-by: Ryan Laurie <iethree@gmail.com>

* MLv2 metadata overhaul, second try

* Stricter linting for MLv2

* Address PR feedback

* bind *card-id* when executing a dashboard subscription so that perm checks work properly

* refactor & address test comment

* fix spacing

* unskip repro that this also fixes

Calling our basis function with just :oss or :ee we don't end up with an
`:argmap` key.

```clojure
uberjar=> (def basis (create-basis :oss))
,#'build.uberjar/basis
uberjar=> (:argmap basis)
nil          ;; nothing extra in oss

uberjar=> (def basis (create-basis :ee))
,#'build.uberjar/basis
uberjar=> (:argmap basis)
{:extra-paths ["enterprise/backend/src"]}

;; and if other aliases are used:
uberjar=> (def basis' (b/create-basis {:project "deps.edn" :aliases [:dev :test]}))
,#'build.uberjar/basis'
uberjar=> (:argmap basis')
{:extra-deps {lambdaisland/deep-diff2 {:mvn/version "2.7.169"},
              methodical/methodical {:mvn/version "0.15.1"},
              io.github.metabase/hawk {:sha
              "45ed36008014f9ac1ea66beb56fb1c4c39f8342b"},
              ....}
 :extra-paths ["dev/src"...]
 :jvm-opts ["-Dmb.run.mode=dev" ...]
 :exec-fn metabase.test-runner/find-and-run-tests-cli}
```

* Fix flaky test and test for a new case

flaky test was annoying. It would timeout after 45 seconds. And
annoyingly, it times out on the _client_, not the server. So we don't
get the proper error message we expect.

And what's happening is that the backend just hooks up the pipe from
it's request to the response. And if no bytes ever come across, we just
twiddle our thumbs until the http request in the test gives up.

The only reliable way i could think of to fix this is to spin up a
webserver that just accepts and then ignores requests (which is what
github did to us in the flaky test).

Had to bump the ring adapter to get the newly exposed `:async-timeout
60000` parameter. By default the webserver in the test has a timeout of
30 seconds, which is shorter than the 45 second timeout for the test's
request to the geojson endpoint.

* add comment

* clean up test server logic

---------

Co-authored-by: Nemanja Glumac <31325167+nemanjaglumac@users.noreply.github.com>

Clean up `metabase.query-processor-test.timezones-test/filter-datetime-by-date-in-timezone-test`  (#28474)

* Get more info about flaky presto test failures

* Include testing context

* Fix flaky test

* Remove redundant `isSettings` prop

* Extract `FormFieldEditorDragContainer`

* Extract `FormFieldEditor` component

* Update field editor UI

* Disable validation in parameters editor

* Fix crash on adding a new parameter

* Add new tests and fix existing

* Remove redundant `overflow: hidden`

* Use shorter flex properties

* Remove redundant `key` prop

* Increase icon size

This reverts commit fbe83eaa.

* fix the flaky test for Bigquery

* adds a limit to make sure we don't have a big charge on CI

* Remove usages of dataset_query.type

* Remove more usages of dataset_query

* Introduce usesSegment and usesMetric methods on Question

* Fix type check

* Fix type

* Get query for UtilQuery

Fixes #19501

Redshift jdbc driver wraps intervals in a special class whereas postgres
simply returns strings. This unwraps the special redshift class into the
same string representation.

Fixes #25378

When using a relative filter with a starting from n units ago on an
aggregation with the same units, we were incorrectly assuming that we
would be working with a field ref at the top level. However, in this
scenario the field is nested such as `[:+ [:field $created_at
{:temporal-unit :month}] [:interval 1 :month]]`. This change makes sure
that we are altering the field-ref.

* Centering tab labels

* adding unit tests

* Switching to userEvent

* update `BulkActions` to new design

* incorporate PR feedback

* update `UndoListing` to match new design for bulk collection actions

* update e2e test to test undo flow

* incorporate PR feedback

* adding hover states, switching icons for buttons

* ChartSettingFieldsPicker unit tests

* fixing background hover state

* using userEvent.click instead of fireEvent

* adjusting e2e tests

* Conversion functions to and from pMBQL

* Appease Kondo

* WIP action data type tests

* DRY up the actions tests

* lots of type tests

* make the tests work

* reorganize tests

* WIP action data type tests

* DRY up the actions tests

* lots of type tests

* make the tests work

* reorganize tests

* reduce some flakes

* better time formatting in action forms

* sort out all the timezones 

* update after rebase

* rebase disaster

These previously passed around `Card`s, now they use `Question`.

The main changes are in `QueryBuilder`'s `handleSave`/`handleCreate`,
and `SaveQuestionModal`.

This also pulls a bunch of logic from `queries/utils/description` that
depends on the internals on `Question.datasetQuery().query` (ie. the
MBQL) into (mostly private) methods on `Question`. The tests for
`description` were moved into `Question.unit.spec.js`.

* move `BulkActionBar`'s styles to `BulkActionBar.styled`

* update `BulkActions` to new design

* update e2e test to not look for `bulk-action-bar` test id

* fix regex for finding `BulkActions` in e2e test

* fix bug where `BulkActions` was appearing then dismissing on page load

* incorporate PR feedback

* Refactor: Simplify code

* Fix custom field parameter lose their settings

Ref: https://github.com/metabase/metabase/issues/22788 this fixes the
issue where custom field parameter seems to be unselected when editing
dashboard a second time. The reason is because how current Metabase lib
is handling expression's MBQL. In FE, we uses `["expression", "name",
null] while in BE, we will normalize it to `["expression" "name"]`.