fix: fast-kill qlds restarts

[dngrtech]

Summary

• switch QLDS systemd stops/restarts to send SIGKILL immediately instead of waiting on SIGTERM timeout
• keep unexpected external SIGKILL/OOM classified as failure by not marking SIGKILL as a successful exit
• bump release/version metadata to 1.13.10

Test Plan

• pytest tests/test_qlds_service_template.py tests/test_systemd_template_render.py tests/test_service_enablement_playbooks.py -q
• Render qlds@.service.j2 and run systemd-analyze verify; no parse errors for new QLDS directives, only expected local missing ExecStart path
• Full pytest currently has an unrelated pre-existing failure: tests/test_script_routes.py::TestValidateScript::test_validate_invalid_python fails when run alone

[github-actions]

PR Review: fast-kill QLDS restarts (v1.13.10)

Strengths

• Well-reasoned design: The comment block in qlds@.service.j2:20-26 clearly explains the three-way constraint: skip the 10s SIGTERM wait (process ignores it), keep restarts fast, and preserve failure detection for OOM/external kills. This is non-obvious systemd behavior explained well in-place.
• Correct decision to omit SuccessExitStatus=SIGKILL: Setting KillSignal=SIGKILL without SuccessExitStatus=SIGKILL is precisely correct. systemctl stop/restart are intentional and systemd won't apply Restart=on-failure for them regardless of exit status. External kill-9/OOM will still be treated as failures and trigger auto-restart — which is the desired behavior.
• Regression guards in the test (test_qlds_service_template.py:28-31): The not in assertions explicitly guard against future introduction of SuccessExitStatus=SIGKILL or a rollback to the old 10s timeout. That's a good use of negative assertions for a subtle behavioral invariant.
• Minimal blast radius: Change touches only the service template, its test, and version bookkeeping. No application code changed.

---

Issues

Critical (Must Fix)

None.

Important (Should Fix)

None.

Minor (Nice to Have)

• TimeoutStopSec=2 rationale could be noted: ansible/templates/qlds@.service.j2:26 — the 2s value is kept to give systemd's cgroup sweep time to collect any child processes after the main process is SIGKILL'd. A brief inline note (or expanding the existing comment) would help future maintainers understand why 0 or 1 wasn't chosen. Not a correctness issue.

• Trailing whitespace removed silently: The old TimeoutStopSec=10 had a trailing space that's now gone. Not a problem, just worth noting it's a silent clean-up bundled in.

---

Assessment

Ready to merge? Yes

Reasoning: The change correctly exploits the systemd contract (Restart=on-failure doesn't apply to deliberate stops) to make config-apply restarts ~8s faster without altering OOM/external-kill behavior. The test covers all key assertions and the comment explains the non-obvious invariant being preserved.