Ensure and document crash resistance and state persistence

[Dominik1999]

Summary

The Guardian must not lose user state in the event of a crash. There should be an industry-standard approach to ensure durability (e.g., WAL, transactional writes, database replication, regular snapshots).

Requirements

• Investigate the current crash recovery behavior
• If state persistence guarantees already exist, document them clearly for operators
• If gaps exist, implement an industry-standard solution
• Consider database replication or regular snapshots as part of the strategy
• Document the recovery process for operators

[zeljkoX]

Postgres-backed deployments (the prod profile, infra directory) already meet the durability bar — RDS handles WAL, automated backups, encryption, and TLS-required connections via RDS Proxy. No additional persistence machinery needed.

Filesystem backend does not meet it (no audit-event persistence, no cross-file atomicity) and should be explicitly documented as dev-only, ideally enforced at startup.

Gap is documentation: no operator runbook for backend-selection guidance and similar topics.

Additional security hardening could be needed for postgres but we will track it with another ticket.