fix(cron): distinguish and clean up script timeouts

[rooroosu]

Summary

• classify no-agent/script deadlines as script timeouts instead of provider failures
• terminate timed-out POSIX script process groups so descendants do not survive
• preserve provider-timeout summaries for true agent/provider failures

Root cause

_summarize_cron_failure_for_delivery treated any error containing timeout as a provider timeout, even when _run_job_script emitted Script timed out after ... for a no-agent job. The script runner also used subprocess.run(timeout=...), which could leave grandchildren alive after the direct child was killed.

Verification

• scripts/run_tests.sh tests/cron/test_cron_script.py tests/cron/test_run_one_job.py tests/cron/test_cron_no_agent.py -q
• 69 tests passed
• git diff --check

Platform behavior

POSIX uses a dedicated session/process group with TERM then KILL fallback. Windows retains compatible direct-child cleanup.

[teknium1]

Thanks for the focused cron cleanup. The premise is confirmed on current origin/main: cron/scheduler.py:75-79 classifies the script runner's Script timed out ... result (cron/scheduler.py:2132-2133) as a provider timeout, and the runner uses a direct-child subprocess.run(..., timeout=...) path at cron/scheduler.py:2100-2108.

Problems

• tests/cron/test_cron_script.py:230 uses os.kill(child_pid, 0) to require the child PID to disappear. A successfully signaled, reparented descendant may remain an unreaped zombie, so this can fail after correct cleanup. Existing process-tree coverage explicitly treats dead-or-zombie as terminated in tests/tools/test_process_registry.py:1928-1952.

Suggested changes

• Make the new descendant assertion zombie-aware, following the existing process-tree test pattern; retain the POSIX-only marker and cleanup guard.

Automated hermes-sweeper review.

[teknium1]

os.kill(pid, 0) also succeeds for an unreaped zombie, so this can fail after successful killpg cleanup when the reparented child has not yet been reaped. Please use a zombie-aware probe, following tests/tools/test_process_registry.py:1928-1952.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #59379 — both fix #59549 with the same two-part change in cron/scheduler.py: check "script timed out" before the generic provider-timeout branch in _summarize_cron_failure_for_delivery, and run the script in its own process group so a timeout kills descendants via os.killpg. #59379 (2026-07-06 04:46) is the earliest-open full-scope fix. Related: #59574 (competing full-scope), #61969 (classification-only subset), #59549 (issue).