Postgres-backed state for multi-replica deployments

[stationeros]

Status

v0.3.0 ships durable database-backed coordination and persistence with SQLite.

Problem

SQLite is fine for local use and pilots, but not for shared production deployment.

Goal

Add Postgres-backed persistence for idempotency records, job state, and execution metadata.

Scope

• storage abstraction for SQLite and Postgres
• migration scripts
• indexed tables for idempotency lookup and job status
• optimistic locking or equivalent write-safety semantics

Acceptance criteria

• service runs against Postgres without API changes
• idempotency replay works across replicas
• job state survives process restart

[snakefood3232]

SQLite works locally but not for shared production, so Postgres-backed state is needed. I'd handle this by adding a storage abstraction layer with SQLAlchemy Core, plus migration scripts for both backends. Can have a PR up in 2-3 days. I've built similar durable job queues with Postgres and idempotency keys.