Sweep stale queue_processes rows on worker startup

[dereuromark]

Summary

Workers that die without graceful shutdown (SIGKILL, OOM, container restart, anything that bypasses the SIGTERM handler) leave their queue_processes row behind with terminate=0. Without intervention these ghost rows accumulate and count toward Queue.maxworkers via QueueProcessesTable::validateCount(), which can refuse to register new workers (Too many workers running) even though the actual processes are long dead.

cleanEndedProcesses() already exists and removes any row whose modified is older than Queue.defaultRequeueTimeout, but it is only invoked from:

• the worker's end-of-loop cleanup, gated by gcprob (default 10%, so ~10 minutes between actual sweeps with one-minute worker turnover);
• a fallback inside the PersistenceFailedException catch block, which only runs after a worker has already failed to start.

Neither path runs before initPid(), so the maxworkers check sees stale rows. This PR moves an unconditional cleanEndedProcesses() call to the top of Processor::run() so every fresh worker sweeps dead siblings before attempting to register.

Reproduction

1. Queue.maxworkers = 2, Queue.defaultRequeueTimeout = 60
2. Start a worker that picks up a long-running job
3. SIGKILL the worker mid-job (or let the container OOM-kill it)
4. The queue_processes row stays with terminate = 0 and modified frozen at the last cycle
5. After 2 such kills the third worker fails to start: Too many workers running (2/2)

After this change, the third worker first sweeps the stale rows and proceeds normally.

Tests

Added to QueueProcessesTableTest:

• testCleanEndedProcessesRemovesStaleRowsOnly — covers the existing method's threshold logic (no behaviour change here, just coverage that was missing).
• testStaleRowsCountTowardMaxWorkersUntilCleaned — regression for the scenario above.

Both pass; existing tests in the file unchanged.

Risk

Low. cleanEndedProcesses() is a single DELETE WHERE modified < threshold, runs in milliseconds, and the threshold is the same value already used by the existing call sites. The only behavioural change is when it runs — once per worker startup vs. occasionally at worker shutdown.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 77.27%. Comparing base (0e8126f) to head (bbf578a).
⚠️ Report is 1 commits behind head on master.
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@             Coverage Diff              @@
##             master     #480      +/-   ##
============================================
+ Coverage     77.25%   77.27%   +0.02%     
  Complexity      955      955              
============================================
  Files            45       45              
  Lines          3214     3213       -1     
============================================
  Hits           2483     2483              
+ Misses          731      730       -1     

☔ 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.