Use docker on CI to speed it up

[fcollonval]

References

Fixes #13600
Closes #11682

Code changes

Building on top of the Docker file we are maintaining for dev (via devcontainer or directly), this switches the CI stack to first build a test docker image and feed it to all most of the tests.

It modifies slightly jupyterlab.browser_check to only install a single web browser based on JLAB_BROWSER_TYPE (that variable was already defined in browser-test.js).

The speed up seen is substantial as creating the docker image take ~8min. But most job execution using it take 3 to 4 min less. It also allows for sharding the integration tests to speed up the overall execution.

Remaining:

• Access data outside the docker to store as artifacts (like for integration or release tests) --> needs to use volume
• Should we speed up the build by caching outside of the Docker image the yarn cache (and potentially the pip cache). This will fasten the build when the yarn.lock changes.
• Should we push a docker image per PR? The issue is that the currently test image is 2.5Go. So if we go that path, I think the only reasonable image would be a minimal python image with JupyterLab installed (assets non-minimized) to allow user testing but not all CI tests.

More optimization:

• Cache Playwright chromium flavor.

User-facing changes

None

Backwards-incompatible changes

None

[jupyterlab-probot]

Thanks for making a pull request to jupyterlab!
To try out this branch on binder, follow this link:

[fcollonval]

First test is encouraging, building the docker image took 6minutes then the 3 tests passing took 3min less (-20% to -35%) than the current job.

[fcollonval]

Sharding of playwright tests is working well. But of course lots of tests are failing likely because the OS config has changed (we are running in the docker). A good side-effect of this PR will be the ability for dev to update the snapshots locally - meaning I'll keep a workflow for it; but we should advice knowledgeable dev to prefer updating just what is needed locally.

[fcollonval]

Just to keep a trace; at the call it was requested to merge the integration tests reports (as this PR splits them). It is supported by Playwright: https://playwright.dev/docs/test-sharding#merging-reports-from-multiple-shards