test: Rebalance parallel test scheduling and trim the slowest tests

[vdusek]

The integration test job grew from ~3.5 min to ~6.3 min when the suite started running across both built-in HTTP transports. Almost none of that was extra work: with 806 tests the suite is only 1432 worker-seconds, which is ~90s on 16 workers. The job took 353s because one worker held the critical path while the other 15 sat idle.

xdist's default load scheduler seeds every worker with a chunk of items_per_node // 4 consecutive tests and never reassigns them. With 806 tests that chunk is 12, and one worker drew the test_build.py block holding all four variants of test_build_delete_and_abort (~87s each), which it then ran back to back: 88s, 170s, 261s, 350s. Every other worker was finished by 170s. The same thing happened before the transport matrix, with 2 variants instead of 4 - so the old 3.5 min was already this one test.

Three changes:

1. --dist worksteal in addopts, so an idle worker takes over tests still queued on a busy one. Measured on the unit suite at identical concurrency: 1558 passed in 40.65s -> 1558 passed in 15.58s. --dist is inert without -n, so single-file runs are unaffected.
2. test_build_delete_and_abort starts both Actor builds before awaiting either. build() returns as soon as the build is queued, so the platform can run them concurrently instead of the test serializing two ~40s builds.
3. test_request_queue_unlock_requests polled list_head until the locked IDs disappeared from it, which list_head does not reflect - so every variant burned the full 30s deadline and then discarded the poll result. It now polls the outcome it cares about, accumulating unlocked_count across unlock_requests() calls until all three locks are accounted for, and asserts that total.

Simulated against the per-test durations reconstructed from the CI logs: 353s -> 100s with change 1 alone, -> 72s with all three.

Change 2 only pays off if the platform builds two versions of the same Actor concurrently. If it serializes them per Actor the test still passes, it just stays at ~90s and the job lands nearer 2 min than 1.5.

✍️ Drafted by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 95.27%. Comparing base (19509a2) to head (23b8ab1).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1037      +/-   ##
==========================================
- Coverage   95.29%   95.27%   -0.02%     
==========================================
  Files          59       59              
  Lines        5501     5501              
==========================================
- Hits         5242     5241       -1     
- Misses        259      260       +1     

Flag	Coverage Δ
integration	92.01% <ø> (-0.19%)	⬇️
unit	87.36% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 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.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.