Integrate ui-profiler, improve dispatch, separate jlpm/yarn

[krassowski]

• Closes Proposal: create a user-facing benchmarking extension #119
• Closes Having other large notebooks open while tab switching #124
• Partially addresses CI is failing on master branch #142
• Closes Fix transpilation #140 (includes, with credit)

Overview

• integrates ui-profiler's Execution Time Benchmark with the existing JupyterLab Benchmark Tests workflow
• all other ui-profiler benchmarks are executed in a new JupyterLab UI Profiler dispatch workflow - example run (:green_circle:)
  • for now those just attach the JSON blob to playwright reports (see below)
  • in a future PR I would like to either:
    • expose the JSON attachments with results as GitHub artifacts for easier download, or
    • upload the JSON results and JupyterLite with ui-profiler installed somewhere (this way we could ensure version match).

Implementation

• calls ui-profiler in playwright via exposed app instance (arguments are just options in JSON format)
• ui-profiler itself is in headless mode (to avoid interfering with the profiling)

Developer-facing changes

New workflows	New dispatch workflow

The new workflow generates an artifact called profiler-report.zip (example):

Which includes playwright reports for both reference and challenger runs:

Each containing an index.html file with a report including the attachment(s):

And each of these reports include attachment with JSON blobs that can be interpreted by copying them into JupyterLab instance with jupyterlab-ui-profiler installed into a file with .profile.json extension in ui-profiler-results top-level directory.

[krassowski]

I believe that the plotly failures are unrelated - I tried increasing the delay a bit and it may helped a bit but I still see:

[testing] › ../jupyterlab/notebook.spec.ts:120:9 › JupyterLab Benchmark › measure largePlotly - 1 
[testing] › ../jupyterlab/notebook.spec.ts:120:9 › JupyterLab Benchmark › measure manyPlotly - 1

[fcollonval]

This is awesome.

I'm fine if this got merge even if we still have error on CI.

[krassowski]

Merging to iterate on it in subsequent PRs.