Fix vaious test flakiness on Windows

[ilevkivskyi]

It looks like there were two sources of flakes:

• In rare cases workers could connect to DB soon, before coordinator configured WAL. I fix this by starting workers after creating metastore. We will probably loose ~5-10ms on this, but I think it is worth it.
• Windows is generally very slow with processes, startup/shutdown sometimes take few seconds. I therefore triple relevant timeouts on Windows.

In addition:

• Handle more parallel workers startup failure scenarios ~gracefully.
• Add a unique build ID to worker status files.
• Add a bit more info in case of a failure.
• Fix one test with unstable error message order.

[ilevkivskyi]

OK, everything points to last flake is caused by Windows being actually slow with processes. Here for example https://github.com/python/mypy/actions/runs/24413561303/job/71326664884?pr=21220#logs the workers didn't print anything for ~3 seconds. I am going to simply scale up all process-related timeouts 3x on Windows.

[ilevkivskyi]

@JukkaL I think this should fix the rest of the test flakiness.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅