[ahernsean]

Fixes #9.

Root cause

The failed web request was not SQLite lock contention. A newer one-shot process had migrated the database while the already-running web service still had older schema support loaded. Archive snapshot creation succeeded, but verification rejected the newer schema.

Changes

• Re-running driving-log install now reloads systemd units and restarts the web service and archive timer when they are active, keeping running code aligned with database migrations.
• Archive creation checks the snapshot schema before publishing a bundle. If the database is newer than the running application, it publishes no archive and reports that the web service must be restarted.
• POST /archives converts that mismatch into a clear 503 response instead of an internal-server-error dialog.
• The speculative read-only snapshot connection change was removed.
• Deployment documentation now records the update/restart behavior.

Tests

• Reproduce a live database schema newer than the running application.
• Verify archive creation rejects it without publishing a .tar.gz bundle.
• Verify the web endpoint returns 503 with restart guidance.
• Verify installation reloads units and conditionally restarts active services.

Validation

• PYTHONPATH=. .venv/bin/pytest -m "not browser" — 86 passed, 1 deselected
• .venv/bin/ruff check driving_log tests
• .venv/bin/ruff format --check driving_log tests
• .venv/bin/mypy driving_log
• git diff --check