* Fixing E2E Cross-version Tests

E2E cross version tests were broken for multiple reasons, primarily due to changes made in UI layout, cypress testing, and data changes. Overall, the basic workflows still work, but subtle click order differences and other small changes made the entire job consistently fail.

To fix this, this PR:
- Provides a function, `parseVersionString`, which computes version information provided a test version string. This information includes the raw string version, the edition (ee or oss), and the major, minor, and patch version of the version string.
- In the same file, `cross-version-helpers.js`, adds constants identifying versions in which certain breaking UI or Cypress changes occurred.
- Finally, a set of exported functions are provided which take a version object and execute conditional logic based on the provided version so that behavior is consistent across Metabase versions.

The versions are computed in `cross-version-source-helpers.js` and `cross-version-target-helpers.js` and are used in js files as shown:

```js
import { visualize } from "e2e/support/helpers";
import {
  fillAreaUnderLineChart,
  newQuestion,
  saveQuestion,
} from "e2e/test/scenarios/cross-version/helpers/cross-version-helpers.js";
```

This then allows logic like the below, where the high level call is made and the right logic is dispatched based on the version information.

```js
it("should create questions", () => {
  cy.signInAsAdmin();

  cy.visit("/question/new");

  newQuestion(version);

  // eslint-disable-next-line no-unscoped-text-selectors -- deprecated usage
  cy.findByText("Orders").click();
})
```

Other smaller changes were also made that created a more universal sequence of UI actions.

Fixes #39229

* clear -> click

* Fixing area icon selection

* Restoring scheduling to align with master

* fe lint

* Dashboard Header FC, using entity, select, and dispatch hooks

* resolving types

* updating other use of getDashboardActions

* PR feedback

* Hack test case to unblock Oracle tests

In metabase.query-processor-test.implicit-joins-test/implicit-joins-with-expressions-test
we calculate an interval by going back from today a number of years to arrive at 2019. On
February 29, this calculation fails with Oracle because there was no February 29 in 2019.
This hack changes 2019 to 2020 and makes Oracle happy.
breakout-on-temporally-bucketed-implicitly-joined-column-inside-source-query-test
had to be adjusted too.

* wip

* SLO works with auth slo handler route

* move slo handling endpoint to /auth/sso/handle_slo

* fix slo redirect url

* SLO works, and the sso-handle-slo for saml is where it belongs

- a ton of cleanup

* fix api/session namespace + add docstrings

* cleaning up logout action

* add slo logout test along with slo response xml

* whitespace + linter

* add docstring

* update exclusions in deps.edn

* un-require metabase-enterprise ns from oss ns

* add docs for how to setup SLO to metabase docs

* docs: clarify that setting up SLO is optional

* move slo logout endpoint into ee code

- removes sso-info defenterprise since it is no longer needed

* use current version of saml20-clj

---------

Co-authored-by: Nick Fitzpatrick <nickfitz.582@gmail.com>

* Don't delete "off" persisted_info records

If they are off, they are "eligible" models for persistence that we do
not want to persist. And to remember that use choice to not persist them
we need to keep the record around. This was handled previously, but
https://github.com/metabase/metabase/pull/39219 was a bit too blunt. We
had a bunch of records in a "creating" state that we weren't removing,
so i deleted them regardless of state. But we want to make sure that the
"off" ones are not deleted.

On stats, a bunch of ones that were set to off are now slated to be
persisted after the prune job deleted the persisted_info records.

* unused local. whoops

* improve error messages

* update error message

* filter notes

* usage insights notes

* Create and use new action `updateUserSetting`

* Add E2E reproduction for #39221

Fixes #39221.

* Make a more versatile and easy-to-use list virtualizer

* use new virtualized list in archive app

* update archive tests

* better mock

* translate object action select strings

* import order fix

* add content from Notion documentation to storybook guidlines for Modal

* clean up examples a bit

* Drop duplicated composes

* De-compose flex

* De-compose flex-full

* De-compose justify-center

* De-compose align-center

* De-compose flex-1-quarter

* De-compose text-centered

* De-compose full

* De-compose relative

* De-compose flex-no-shrink

* De-compose block

* De-compose inline-block

* De-compose flex-row

* De-compose ml-auto

* De-compose mr-auto

* De-compose full-height

* De-compose absolute

* De-compose top

* De-compose align-self-start

* De-compose no-decoration

* De-compose flex-half

* De-compose cursor-pointer

* De-compose flex-aligh-right

* Add local-global styles

* Add local-global styles

* Fix missed rules

* Fix vendors styles

* Revert emotion refactors

* Fix vendor styles

* Fix eslint

* Fix e2e tests

* Fix review comments

* Fix

* Code review fixes

* Hotfix after merge

Fixes #39053.

* Show unsaved changes warning for native questions only when the query has been changed

* Drop legacyQuery usage

* Add test for new behavior

* Update pathnames for running questions
- required to make native_subquery.cy.spec.js pass - this test suite uses native queries referencing other questions by id instead of a full slug, and upon opening such question the id in the query automatically changes to the full slug, and the URL gets a hash

Removing a trailing `/`

* Add new user-specific setting `last-used-database-id`

* Update user settings when user selects a database for a native query

* Remember the last used database in the native query editor

* Fix type error

* Set database id when initializing new native question

* Persist the database only for "New > Native query"

* Persist last used database for native models

* Mark `last-used-database-id` setting as potentially `undefined`

* Reduce diff

* Remove unused prop from QB

* Lift dispatch logic up to the `View` component

* Fix type errors

* Make this setting work for all authenticated users

* Add initial E2E tests

* Add initial backend unit test

* Expand the backend unit test to check for user-local only property

* Fix E2E test and make it more explicit

* Add mongo-related E2E test

* Fix typo

* Add permission-related E2E tests

- Adds reproduction for #39053

* Make sure no redundant requests are made

This commit ensures no redundant requests are made when selecting
the same, i.e. the previously selected, database.

* Add E2E tests related to model actions

* Clean up

* Extract persist database logic into an action

* Update the setting name as suggested in the code review

* Make the setting description a bit more explicit

* Add one more permissions-based E2E test

* Store postgres database name in a variable

* Keep result-metadata of native queries if possible

Part of #37009.

* Fix typo

* Add a fallback for when resultMetadata is not available

* Revert "Add a fallback for when resultMetadata is not available"

This reverts commit 5ad3a8b4081ed192267621b716b1634d304f470f.

* Move resultsMetadata logic into getSubmittableQuestion

* Add a fallback for when resultMetadata is not available

* Add explanatory comment

* Guard against missing resultsMetadata

* Add repro for #35039

* Simplify the test

* Add an explanatory comment

* Fix test

* Rename promise to runQuestionPromise

---------

Co-authored-by: Tamás Benkő <tamas@metabase.com>

* Remove unused props to avoid confusion

* Fix incorrect official collection icon in collection view

* Fix hardcoded SVG icon color

* Remove unused prop

We prune persisted info records if they meet any of the following
criteria:

```clojure
(or (contains? (persisted-info/prunable-states) current-state)
    (:archived card-info)
    (not (:dataset card-info)))
```

But we only deleted the record when:

```clojure
(when (= "deletable" current-state)
  (t2/delete! PersistedInfo :id (:id persisted-info)))
```

So any records that were in a "creating" state (persist a model, but
before it first gets persisted, make it not a model, or archive the
underlying model), we constantly pruned them but never removed the
persisted info record.

Leading to task results like:

```javascript
{"success": 21, "error": 0, "skipped": 0}
```

Because 21 things were queued up for pruning, were attempted to be
pruned, but the persisted info record never removed.

* Drop report_card.dataset [WIP] [ci skip]

* Drop report_card.dataset

* Some test fixes

* More test fixes

* All OSS tests should be fixed now?

* Remove check for :dataset key from snake-hating-map

* Remove NOCOMMIT stuff

* Remove unused namespaces

* PR feedback

* Migrate dataset attribute in autocomplete suggestions (#39054)

* FE - Migrate `dataset: true` to enum value in Bookmarks (#39056)

* Fix Card["type"] - Bookmark["card_type"] mapping and add an extra assertion

* Fix card-type check out of raw query

* Fix the test

* Fix tests after merge

* Log body when unexpected response code

* Get logging in CI

* Relax is_upload schema for latest mariadb

---------

Co-authored-by: Kamil Mielnik <kamil@kamilmielnik.com>
Co-authored-by: Case Nelson <case@metabase.com>
Co-authored-by: Alexander Polyankin <alexander.polyankin@metabase.com>

This PR updates the logic of skipping what cards are displayed in a pulse email by ensuring the type of the card being processed is a `:card` in addition to having no results.

This PR narrows down the matching pattern for the milestone reminder to
only the descriptions that explicitly contain one of the closing keywords
followed by the issue (number).

To verify that the proposed solution works, please check:
- https://regexr.com/7sk94

Also see the list of the closing keywords as defined by GitHub.
https://docs.github.com/en/get-started/writing-on-github/working-with-advanced-formatting/using-keywords-in-issues-and-pull-requests



* Make the pattern case insensitive

Co-authored-by: Nicolò Pretto <info@npretto.com>

* Fixing bad pulses when bad params exist

Previously, when a dashboard parameter was removed that was depended upon by a subscription, the subscription went into a broken state that was unfixable (the removed param was still attached to the notification but not visible in the UI for editing). Rather than remove that parameter, which could potentially lead to data leaks, we just archive the pulse.

Now, when a dashboard is saved that would break a pulse/subscription, we archive the pulse and send an email the the dashboard and pulse creators to notify them that the pulse has been removed. The email also contains which params were removed as well as a list of who will no longer be receiving the messages.

One code consideration when writing the query logic to determine the blast radius of the bad pulse was whether to write a large join query then reorganize the results or do N+1 queries starting from the pulse id (The logic requires hitting all of pulse, pulse channel, pulse channel recipients, and user tables). I started with a full join query along the lines of what is shown below, but opted for the N+1 query as broken pulses should be a very rare occurrence so readability was preferred over performance.

```clojure
  (let [parameter-ids ["a3d043d5"]]
    (when (seq parameter-ids)
      (t2/query
        (sql/format
          {:with   [[:params {:select [[:id]
                                       [[:raw "json_array_elements(parameters::json) ->> 'id'"]
                                        :parameter_id]]
                              :from   :pulse}]
                    [:pulse_ids {:select-distinct [[:id :pulse_id]]
                                 :where           [:in :parameter_id parameter-ids]
                                 :from            :params}]]
           :select [[:p.id :pulse_id]
                    [:pc.id :pulse_channel_id]
                    [:pcr.id :pulse_channel_recipient_id]
                    [:creator.first_name :creator_first_name]
                    [:creator.last_name :creator_last_name]
                    [:creator.email :creator_email]
                    [:recipient.first_name :recipient_first_name]
                    [:recipient.last_name :recipient_last_name]
                    [:recipient.email :recipient_email]]
           :from   [[:pulse :p]]
           :join   [:pulse_ids [:= :p.id :pulse_ids.pulse_id]
                    [:pulse_channel :pc] [:= :pc.pulse_id :p.id]
                    [:pulse_channel_recipient :pcr] [:= :pcr.pulse_channel_id :pc.id]
                    ;; Get the pulse creator
                    [:core_user :creator] [:= :creator.id :p.creator_id]
                    ;; Get the pulse recipient
                    [:core_user :recipient] [:= :recipient.id :pcr.user_id]]}))))
```

* Fixing inconsistent test results by adding a timeout and expecting 2 messages.

* Fixing NPEs in tests, but not sure why we are not getting results sometimes.

* Trying to fix 500 in dashboard PUT

* Tests run ok locally against mongodb

* Fixing unit tests

* fixing template up

* fixing unit tests

* Adding refined tests to try to detect what is going on with failing tests

* Formatting

* Tracking down test failures and updating email template

* Removing db-specific query

* Simplifying broken pulse logic to not get all pulses twice.

* A couple small simplifications to the logic.

* Adding slack messaging

* Making pulse checks conditional on parameter updates