Show test failures of ci-meson as annotations

[user202729]

Follow-up to #36938 and #37738 , this PR shows the test failures of ci-meson.yml workflow as GitHub annotations, which has the advantage that you don't need to scroll through the whole log to view the failed tests.

(actually I just search for Failed examples, but this is not foolproof)

Previously only build.yml got this feature.

There's a disadvantage, then the file and line numbers are no longer shown in the downloaded log file, instead only [error] is shown (but this is trivially fixable by printing the format in both old and new ways when --format github is passed). I think if everyone views the annotations instead of reading the raw log anyway, it shouldn't matter too much.

📝 Checklist

• The title is concise and informative.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes. (no functionality change)
• I have updated the documentation and checked the documentation preview. (no documentation change)

⌛ Dependencies

[github-actions]

Documentation preview for this PR (built with commit 763521d; changes) is ready! 🎉
This preview will update shortly after each push to this PR.

[user202729]

Looks like it's working. (before the change, the highlighted part would always be test-long or test-new instead.)

[saraedum]

Looks good to me. But maybe @tobiasdiez (or somebody else) could sign off on this one?

[tobiasdiez]

Thanks!

This has the slight disadvantage that now genuine errors are annotated a few times, but for this the github interface just doesn't give a better solution atm.

[tobiasdiez]

(concerning the two failing tests, can they not be fixed with the alarm-test-helper that you have introduced recently @user202729 )

[user202729]

The alarm test thing is just a convenient wrapper over alarm, but does almost exactly the same thing (plus ensuring the time from interruption to exception thrown isn't too long).

I don't know what's the cause of the test failure, but it says "time out" means even after the "user" press ctrl-C, wait for ≈ 10 minutes, the test still doesn't finish. Of course I've no idea why (last time I check signals cannot be ignored), but that would genuinely be a test failure.

I guess we can try to set two alarms somehow (simulate "user press ctrl-C again if it doesn't work the first time"). If tests still fail at least we get a slightly better idea why they do (more likely deadlock, less likely signal ignored/erroneously caught).

p/s: actually I did get the (supposedly) lll (low-level lock?) deadlock once locally on my machine, but never got around to debug it.