Fix flaky backend CI worker timeout

[JSv4]

Summary

• isolate the missing-user message vector-search test from database, permission, and embedder fixtures
• force the intended User.DoesNotExist branch with mocks in a SimpleTestCase
• use pytest-timeout's signal mode so xdist reports timed-out tests instead of losing a worker and hanging the controller

Root cause

The Backend CI failure observed in #2170 was unrelated to that PR's dependency update. Across repeated failures, an xdist worker died exactly 600 seconds after the second 41% progress marker. A CI-equivalent coverage trace matched that timer start to MessageVectorSearchTest::test_message_search_nonexistent_user within 0.4 ms and on the same worker.

The test exercised a full database/permissions/embedding fixture merely to cover the missing-user branch. When the existing thread-based timeout fired, pytest-timeout called os._exit(1), causing xdist to report node down: Not properly terminated and wait until the outer 100-minute timeout.

Impact

The branch test is now deterministic and dependency-free. If another backend test genuinely hangs, CI will report its node ID and stack as a normal pytest failure instead of silently losing the worker.

Validation

• isolated missing-user test: 1 passed
• complete conversation-search module under xdist and signal timeout: 82 passed in 20.12s
• controlled 50 ms timeout: reported the exact test and exited xdist cleanly without node down
• black --check opencontractserver/tests/test_conversation_search.py
• flake8 opencontractserver/tests/test_conversation_search.py
• git diff --check

A clean full backend run reached 26% without failures before it was stopped to publish the fix.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!