fix: sync database storage incrementally

[yanhao98]

Summary

• replace the database storage backend's delete-all/reinsert-all snapshot writes with incremental synchronization
• preserve unchanged database rows and IDs
• update only changed JSON payloads, insert new keys, and delete keys missing from the new snapshot
• add regression tests covering reordered snapshots, changed rows, inserted rows, removed rows, duplicate-key rollback/redaction, and the auth-key id to key_id mapping

Why

The previous implementation rewrote every account on every save. With PostgreSQL and large token/JSON fields this creates large numbers of dead TOAST tuples. A production table with roughly 2,782 live accounts grew to approximately 109 GB, of which approximately 108 GB was TOAST, while the compacted live relation was only tens of MB.

Closes #355

Test plan

uv run --with pytest pytest -q test/test_database_storage.py

Result:

6 passed

The repository-wide pytest invocation currently also collects integration scripts that require a running server and configured auth; those pre-existing tests fail locally with connection-refused/401 errors and are unrelated to this storage change.