fix(cron): keep active manual runs heartbeating

[qdivan]

What does this PR do?

Keeps the calling turn's activity heartbeat alive for the full duration of an inline manual cron run.

The heartbeat currently stops after a hard-coded six-hour ceiling even when run_one_job is still active. Once it stops, the parent inactivity watchdog can terminate a legitimate long-running job at a deterministic wall-clock threshold. Elapsed time alone is not evidence that the child is wedged: persistent gateway sessions already dispatch manual runs through the async-delegation rail, while synchronous fallbacks remain governed by the cron run's own timeout/cancellation policy (including an explicit unlimited setting).

This removes the independent heartbeat ceiling and leaves the heartbeat tied to the actual run_one_job lifetime. The existing finally block still stops and joins the heartbeat thread as soon as the job exits.

Related Issue

No linked issue; this is a self-found regression in the manual-run heartbeat introduced by #76675. It is distinct from the opt-in per-job caps proposed in #45809.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✅ Tests (adding or improving test coverage)

Changes Made

• tools/cronjob_tools.py: remove the elapsed-time gate that drops activity while an inline cron run is still executing.
• tests/tools/test_cronjob_run_immediate.py: replace the old ceiling contract with a deterministic monotonic-clock regression proving heartbeats continue beyond six hours until completion.

How to Test

1. Run scripts/run_tests.sh tests/tools/test_cronjob_run_immediate.py tests/tools/test_cronjob_run_background.py tests/tools/test_cronjob_tools.py -q (94 passed).
2. Run scripts/run_tests.sh tests/cron/ -q (873 passed, 1 Windows-only test skipped on Linux).
3. Run uv run --with ruff ruff check tools/cronjob_tools.py tests/tools/test_cronjob_run_immediate.py.

The focused regression was verified RED before the production change: the old code logged that the ceiling was reached, delivered no heartbeat, and failed while the fake job remained active. It passes after this change.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix
• I've run the entire tests/ suite (focused tool and cron suites above are green)
• I've added tests for my changes
• I've tested on Linux

Documentation & Housekeeping

• Documentation update — N/A; no user-facing contract or config changed
• cli-config.yaml.example update — N/A; no config keys changed
• CONTRIBUTING.md / AGENTS.md update — N/A; no workflow changed
• Cross-platform impact considered — stdlib threading/time behavior is platform-neutral
• Tool descriptions/schemas update — N/A; tool schema is unchanged

[Enough1122]

AI code review — automated review for reference; please use your judgment.

• tools/cronjob_tools.py:835-845 — removing the ceiling entirely resurrects the pre-[Bug]: cronjob action='run' blocks the calling turn synchronously — inactivity watchdog kills the parent agent at 1800s #76502 hazard its comment documented: the heartbeat proves the thread is alive, not that the job is progressing, so a wedged run_one_job under HERMES_CRON_TIMEOUT=0 now masks the gateway watchdog forever (the new test literally pins a sleeping job being heartbeated). Please reintroduce a backstop — a much higher hard cap, or a progress-aware signal (e.g., stop heartbeating if no child-side activity marker advances for N intervals) — or document why unlimited masking is now acceptable.
• tools/cronjob_tools.py:27-31 (removed comment) — operators who currently depend on the 6h reap to clean up hung manual runs will see turns that never die after this lands; this belongs in the changelog with the recommended mitigation ("set HERMES_CRON_TIMEOUT instead of relying on the ceiling").
• tests/tools/test_cronjob_run_immediate.py:231-262 — the flipped test cleanly pins the new contract (elapsed time alone must not drop protection); add its missing counterpart once a backstop exists: a wedged job with unlimited child timeout still gets reaped by something, and the old "heartbeat stops at ceiling" path has a test asserting whatever replaces it.
• nit: with the ceiling gone, the loop's logger.warning import/usage may be unused now — check lint; also consider logging at INFO every hour while heartbeating so ultra-long legitimate runs remain visible in logs.