192 Job schedule completion updates

• Status: Accepted
• Date: 2026-02-07
• Context:
  • Motivation: enforce ERD job_schedule completion semantics (next_run_at + lock cleanup) for indexer jobs.
  • Constraints: stored-procedure-only runtime access, constant error messages, versioned procs with stable wrappers.
• Decision:
  • Add job_schedule_mark_completed_v1 and job_run_*_v2 wrappers to update last_run_at, next_run_at, and clear locks on both success and failure.
  • Keep job_run_reputation_rollup signature stable by mapping window_key to job_key in-proc.
  • Alternatives considered: update next_run_at in job_claim_next (rejected; ERD mandates update on completion) and update schedule in app runner (rejected; DB is SSOT).
• Consequences:
  • Positive outcomes: job_schedule rows now reflect completion cadence with jitter and lock cleanup per ERD.
  • Risks or trade-offs: if job_schedule_mark_completed_v1 fails, job errors are surfaced as schedule update failures.
• Follow-up:
  • Verify any future job runner wiring calls job_run_* wrappers (not versioned functions directly).
  • Review checkpoint: confirm Phase 9 checklist remains aligned with ERD job cadence rules.
• Test coverage summary:
  • Added a stored-proc test asserting job_run_retention_purge updates schedule timestamps and clears locks.
• Observability updates:
  • None (database-only change).
• Risk & rollback plan:
  • Roll back by reverting migration 0084 and restoring job_run_* wrappers to v1.
• Dependency rationale:
  • No new dependencies. Alternatives considered: not applicable.