Harden CI database migration deploy retries

[kevinhermawan]

Summary

CI deploys are intermittently failing during prisma migrate deploy against orchestration/mediapulse databases, then succeeding on retry. We should harden the deployment workflow so transient DB/network issues do not require manual reruns while keeping migration execution safe and deterministic.

Scope

• In scope:
  • Add a dedicated migration stage in CI that runs before deploy jobs.
  • Add bounded retry with backoff for transient migration failures.
  • Ensure migrations are not executed concurrently from multiple deploy jobs.
  • Document expected behavior and failure conditions in workflow comments.
• Out of scope:
  • Changing Prisma schema or migration SQL content.
  • Reworking application runtime DB connection logic.

Acceptance criteria

• CI has a single migration job that runs before app/agent deploy jobs.
• Migration step retries transient failures (e.g. network hiccups) with bounded attempts and delay.
• Deploy workflows no longer trigger duplicate/concurrent migration execution paths.
• Workflow fails clearly after retry budget is exhausted.
• At least one successful pipeline run demonstrates migration + deploy flow without manual rerun.

Dependencies

• Depends on: none
• Blocks: none

Notes

Observed symptom: logs show prisma migrate deploy failing intermittently in CI and succeeding when the workflow is retried manually.