Alert monitors and ADR/PRD closeout

[ncolesummers]

Summary

Define the first alert monitors for Loopworks operational telemetry and close out the ADR 0012 / PRD documentation follow-through after preview telemetry validation succeeds.

Deliverables

• Define alert thresholds for stuck runs, failed webhooks, approval age, and repeated validation failures.
• Implement or document Axiom monitors and notification routes with dry-run or test evidence.
• Record the alert rationale and escalation expectations in observability docs.
• Resolve PRD open question 6 by linking to ADR 0012.
• Move ADR 0012 from Proposed to Accepted only after the preview deployment validates Axiom telemetry end to end.

Acceptance Criteria

• Focused tests, dry-run checks, or reproducible monitor validation are defined before implementation, with a clear red or missing-state record.
• Alert thresholds are documented with rationale and expected operator action.
• Alert routes are tested, dry-run verified, or explicitly documented as pending external setup.
• PRD open question 6 links to ADR 0012 instead of remaining open.
• ADR 0012 is accepted only after the preview validation issue confirms traces, metrics, logs, and trace-id correlation in Axiom.
• Docs changes follow the docs/ADR guides; broad observability changes finish with bun run validate.

Notes

• This issue should run after the OTel/Axiom preview validation issue.
• Keep threshold choices practical for a solo-operator control plane and adjust later from observed production volume.
• Do not close the backend decision history; preserve the Grafana fallback in ADR 0012.

Refs #21
Refs ADR 0012