fix(tests): repair default test failures

[smithfabian]

Fixes a few test found from a fresh fork + clone setup. The fixes make the current default test suite behave consistently for new contributors.

The failing tests were:

1. tests/test_accuracy_benchmark.py::TestAccuracyBenchmarkRequest::test_all_valid_benchmarks
2. tests/test_admin_api_key.py::TestListModelsSettings::test_list_models_includes_all_model_settings_fields
3. tests/test_admin_profiles_api.py::test_all_model_settings_fields_classified

Plus two additional failures found while validating the test suite:

4. tests/integration/test_e2e_streaming.py
5. tests/test_boundary_snapshot_store.py::TestBoundarySnapshotSSDStore::test_cleanup_all_drains_queue

This also fixes:

6. A fresh dev install dependency failure where tests using FastAPI with form/file handling can fail during app import.

Changes

• Derives the accuracy benchmark expected count using len(VALID_BENCHMARKS) instead of a stale hard-coded integer. Fixes failing test 1.
• Include preserve_thinking and turboquant_skip_last in the admin model settings response. Fixes failing test 2.
• Classify preserve_thinking as a universal profile field in UNIVERSAL_PROFILE_FIELDS. Fixes failing test 3.
• Add MockEnginePool.get_entry() for mocked streaming tests. Fixes validation gap 4.
• Fix a BoundarySnapshotSSDStore.cleanup_all() race where cleanup could miss an in-flight background write. Fixes validation gap 5.
• Add missing python-multipart dev dependency needed by FastAPI form/file routes in the test suite. Fixes failure 6, for fresh pip install -e ".[dev]".

Test validation

pytest tests/test_accuracy_benchmark.py::TestAccuracyBenchmarkRequest::test_all_valid_benchmarks \
  tests/test_admin_api_key.py::TestListModelsSettings::test_list_models_includes_all_model_settings_fields \
  tests/test_admin_profiles_api.py::test_all_model_settings_fields_classified
# result: 3 passed

pytest tests/integration/test_e2e_streaming.py --override-ini=addopts=
# result: 37 passed

pytest tests/test_boundary_snapshot_store.py::TestBoundarySnapshotSSDStore::test_cleanup_all_drains_queue
# result: 1 passed

pytest
# result: 3837 passed, 18 skipped, 54 deselected

[Copilot]

Pull request overview

This PR fixes several “fresh clone” test failures and improves default test-suite consistency for new contributors by aligning tests with current constants, completing admin model-settings responses, strengthening mocked streaming infrastructure, addressing a boundary snapshot cleanup race, and ensuring required FastAPI multipart support is present in dev installs.

Changes:

• Make accuracy benchmark test expectations derive from VALID_BENCHMARKS instead of a hard-coded count.
• Expose missing model settings fields in the admin “list models” response and classify preserve_thinking as a universal profile field.
• Fix streaming test mocking gaps, harden BoundarySnapshotSSDStore.cleanup_all() against in-flight writes, and add python-multipart to dev dependencies.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/test_accuracy_benchmark.py	Removes stale hard-coded benchmark count by asserting against len(VALID_BENCHMARKS).
tests/integration/test_e2e_streaming.py	Extends the mock engine pool with get_entry() to satisfy server code paths during streaming tests.
pyproject.toml	Adds python-multipart to dev dependency sets to prevent FastAPI import-time failures for form/file routes in tests.
omlx/model_profiles.py	Adds preserve_thinking to UNIVERSAL_PROFILE_FIELDS for correct profile/template field classification.
omlx/cache/boundary_snapshot_store.py	Ensures queue tasks are marked done and uses join() to avoid cleanup races with in-flight background writes.
omlx/admin/routes.py	Includes preserve_thinking and turboquant_skip_last in the admin model settings response payload.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.