[ehippy]

Rebase of #2 (originally by @LAIM69420) onto current main. The curation-kind refactor landed on main after this PR was opened (bug_sweep was collapsed into overseer, see 73a0eda); the rebase carried the fix forward onto the overseer version of the test via context matching, no manual edits needed.

What changed

• stub the detached curation coroutine in the API routing test
• wait until the scheduled stub has actually run before the test exits

Root cause

test_curate_endpoint_requires_auth_and_starts_in_background launched the real curation pass with asyncio.create_task and returned without joining it. The following test's fixture could then clear the shared SQLite database while that detached task still held or used another session, intermittently producing sqlite3.OperationalError: database is locked at the next test's setup.

Note: main separately picked up WAL mode + a busy_timeout at the engine level (ca91d67) since this PR was opened, which addresses the same symptom from a different angle. That's a good belt-and-suspenders complement, not a replacement — it doesn't make this test's background task deterministic, it just makes lock contention less likely to surface as a hard failure while the race still exists.

Validation

• pytest tests/integration/test_api.py -q — 11 passed
• pytest -q four consecutive times — 509 passed each run
• ruff check tests/integration/test_api.py — clean

Closes #2.