fix: align managed agent shutdown timing

[rung]

Refs cicd-sensor/cicd-sensor#143
Pairs with cicd-sensor/cicd-sensor#148.

Why

The Action-managed Agent unit used TimeoutStopSec=5s, while the Agent needs a longer grace period to flush manager output during shutdown. systemd could therefore send SIGKILL before the Agent finished its own drain.

The Action owns both the Agent command line and its transient systemd unit, so those deadlines should be configured together.

What

• pass --shutdown-grace 20s to Action-managed Agents
• increase the Agent unit's TimeoutStopSec to 30s
• keep the Docker proxy timeout at 5s
• test that the supervisor timeout remains greater than the Agent grace period
• update the checked-in dist/main bundle

Verification

• npm run lint
• npm test
• npm run check-dist
• 10-job GitHub-hosted verification using the exact Agent PR #148 and Action commits:
  • GitHub Actions: 10/10 succeeded
  • Manager summary logs: 10/10
  • Manager detection logs: 10/10
  • runtime event logs exactly matched events_total: 9/10

The remaining incomplete runtime tail shows that timeout alignment alone does not guarantee every streaming record under concurrent teardown. This PR fixes the Action's supervisor deadline mismatch; Agent-side project-result/finalization behavior remains owned by #148.

[rung]

CI note: the source checks and all 23 tests pass. The job currently fails only at npm audit because the unchanged main lockfile now reports brace-expansion (high) and undici (moderate) advisories. I am keeping that dependency/security update separate from this shutdown-timing change.