* Update & tag a test

* Run tests also on TS files

* Update mongo test

* Explicitly decide whether to upload or not

* Upload in the main e2e workflow, but skip uploading in stress-test workflow

* Do not record runs in the stress-test workflow

* Add Replay API key

Even though we're not using it in this workflow. 

* Omit env var from the stress-test workflow

Resolves #38301

* Upgrade `actions/download-artifact` action to v4

* Upgrade `actions/upload-artifact` action to v4

Co-authored-by: Vladimir Zaytsev <vladimir@trunk.io>

* Remove Cypress cloud integration

* Remove custom Chromium

From now on, we'll use `replay-chromium` browser.

* Start using ReplayIO for all E2E test runs

* Use the default Chrome for Percy visual tests

* Mark flaky tests

* Revert "Improve external filter checks (#35740)"

This reverts commit 8ef7eb59.

* Revert "Make Chromatic workflow 3rd party friendly (#35708)"

This reverts commit 931029e0.

* Revert "Make E2E workflow 3rd party friendly (#35643)"

This reverts commit b547eb7e.

* Revert "PoC External workflow run trigger (#34951)"

This reverts commit df16fc89.

* Improve and unify `external-filter` checks

* Refine E2E jobs conditionals

* add special flaky group

* tag tests with flakes over 10%

* mark more flaky jobs

* Stop caching the uberjar for E2E tests

* Bring back SHA1 in an uberjar name

* Adjust the download uberjar action

* Fix uberjar name for percy

* Separate Replay E2E tests from the main workflow

* Fix workflow conditionals

* Move E2E download uberjar logic into a composite action

* Check out only the last 20 commits

* Always use the correct, last commit in a branch

This helps us avoid the slippery slope in which GitHub creates a new
merge commit, that doesn't exist in the repo, under our feet.

* Adapt the stress-test workflow to use new download action

* Increase the number of commits to search through

* Update .github/actions/e2e-download-uberjar/action.yml

Co-authored-by: Ryan Laurie <30528226+iethree@users.noreply.github.com>

* Log full artifact name

* Correctly download artifact from the current build

---------

Co-authored-by: Ryan Laurie <30528226+iethree@users.noreply.github.com>

No need to repeat the same line for every single **step**
when all steps can read from the **job** context.

* Conditionally start Docker containers for E2E tests

* Check out before initializing Docker containers

* Use official action to log in to DockerHub

* Evaluate conditional expressions

* Try shell conditionals

* Fix errors due to unset vars

* Run either Mongo or Pg+MySQL

This was causing scheduled runs to fail due to a
missing closing quotes in a string.

[ci skip]

* Move browser-related Cypress prep steps to a composite action

* Fix browser paths

* Run E2E using ReplayIO daily

Increase the frequency from weekly to daily.
[ci skip]

* Use the latest replayio/action-upload version

[ci skip]

* Tag Mongo-dependent E2E tests

* Run mongo tests only if `CYPRESS_QA_DB_MONGO` is `true`

* Adjust E2E workflow to run mongo tests separately

* Guard against missing mongo env var

* Fix workflow

* Fix qa snapshots

* Tag additional Mongo E2E tests

* Omit tests tagged with `@mongo` in regular runs

The original attempt was #33777.

As revealed in #33799, the issue hasn't been fully fixed.
Take a look at this run: https://github.com/metabase/metabase/actions/runs/6110515399

The skipped test names are now in this format: e2e-tests-collectionsslow-ee
Let's see how do we create a name:

name: e2e-tests-${{ matrix.folder }}${{ matrix.context }}-${{ matrix.edition }}
So it's obvious that both folder and context are included, when in reality we need only one of the two.

The flaw in the logic was trying to remove the runner from the matrix. At least ONE matrix parameter needs to be different in the slow context. We achieve this by setting the runner to ubuntu-latest instead of ubuntu-22.04. This should now finally fix the names.

[ci skip]

* Update additional context for E2E stubbed/skipped tests

* Include two new E2E groups to stubbed/skipped E2E tests

* Run slow E2E tests using faster runner

* Tag slow tests

* Add additional x-ray tests to the slow batch

This log shows the output of the files changed:
https://github.com/metabase/metabase/actions/runs/6087912385/job/16537680230#step:3:602

```
Changes output set to ["shared_sources","frontend_sources","frontend_specs","frontend_all","backend_sources","backend_specs","backend_all","sources","e2e_specs","e2e_all","codeql","i18n","visualizations"]
```

Yet, the `download-uberjar` job ran, and the `build` never ran.
This resulted in E2E test failures due to missing FE changes.

The fault in the logic introduced in #33190 was thinking that
`needs.files-changed.outputs.e2e_specs == 'true'` means the only output,
when in reality it means one of many outputs.

This PR should fix this by making sure that `needs.files-changed.outputs.e2e_all != 'true'`
is also respected. This translates to - if there were no source code changes.

[ci skip]

* PoC: Download previously saved uberjar for E2E workflow

* Separate `e2e_specs` and `e2e_all` file path triggers

* Fix file path result oputput

* Refine E2E tests conditional trigger

* Further refine E2E tests conditional trigger

* Download the uberjar directly where it's needed

* Work around inability to dynamically define job ouputs

* First obtain the artifact, then run anything else

* Map job outputs

https://docs.github.com/en/actions/using-jobs/defining-outputs-for-jobs

* Apply changes to visual regression tests

* Final logic cleanup

* Work around the API rate limit

* Split `admin` E2E group and merge it with `auditing`

* Run new `admin-2` group in CI

* Move dashboard bookmarks spec to `organization`

* Move dashboard filter related specs to appropriate group

* Create new `dashboard-cards` E2E group

based on https://github.com/orgs/community/discussions/26303 and my experience, workflows with `always()` will not cancel even if a new run of the same workflow appears with concurrency and cancel-in-progress:true. Official documentation suggests using `!cancelled()` instead

* Install Chrome immediately after checkout

* Switch to Chromium 111.0.5563.146