## Summary
- Remove the 600Mi memory limit on the local Postgres Deployment. With `shared_buffers=384MB` + `max_connections=300` the pod was being repeatedly OOMKilled under real workloads, leaving clients with intermittent `connect: connection refused`. Keep the 400Mi request for scheduling.
- Add `document` to the seeded database list in `initdb.sh`.
## Test plan
- [x] `kubectl --context kind-unbound apply -f k8s/infra/postgres.yaml` — rollout succeeds
- [x] Postgres pod stable (0 restarts) after apply
- [ ] On a fresh cluster (`./setup`), verify the `document` database is created (requires empty `data/postgres/`)
Reviewed-on: #259
argoyle
73ccafab68