fix benchmark result posting

[hakkelt]

In this PR, I try to fix this problem by dividing the benchmarking into two phases:

1. Run the benchmark, save the results to a file, and upload it as an artifact. -> It does not need any special permissions.
2. Download the artifact and post the content as a comment. -> This gets triggered by the first phase via "workflow_run", and my hope is that it is possible to selectively set write permission to this job only.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.78%. Comparing base (83c73c9) to head (ab571a6).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #159      +/-   ##
==========================================
- Coverage   94.79%   94.78%   -0.01%     
==========================================
  Files          79       79              
  Lines        2498     2494       -4     
==========================================
- Hits         2368     2364       -4     
  Misses        130      130              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hakkelt]

To test this PR, post_benchmark_comment.yml needs to be on master already... So I'll play around a bit in some playground repositories, then I'll report if it works.

[hakkelt]

I opened a PR on AirspeedVelocity (MilesCranmer/AirspeedVelocity.jl#142) that adds this two-stage benchmarking workflow.

@lostella For now, there are three options, I think:

• Merge the other benchmarking PR that saves the results into the job summary (Benchmarking results as job summary #158)
• Merge this PR that uses my fork and posts comments safely
• Wait until my PR on AirspeedVelocity gets merged, then update this PR to use the new version

Regarding the automatic detection of performance regressions, I considered it and concluded that making the job fail might not be the best solution. I opened another PR on AirspeedVelocity (MilesCranmer/AirspeedVelocity.jl#140) that adds emojis to its result table to mark significant speedups and slowdowns (actually, I just re-heated the original author's outdated PR). If it gets merged, I'm considering opening another one that adds a notice before the collapsible table in the posted comment, warning about the negative performance regressions.