test(backend): reuse workflow selector discovery

[tianmind-studio]

What changed and why

Reuse one module-scoped workflow-selector discovery across the contract tests instead of reloading the selector and rescanning every backend unit test in each test. This keeps the Windows fast-unit lane below its per-file timing budget without changing the contract assertions. Closes #9729.

Product invariants affected

none

How it was verified

• Reproduced the baseline twice on Windows: all 19 tests passed, but the fast-unit duration guard rejected the file at about 0.14s and 0.12s against the 0.12s budget.
• Repeated the strict backend/test.sh run five times after the change. All 19 tests passed each time, with total runs of 0.57s, 0.47s, 0.54s, 0.52s, and 0.60s and no duration warning.
• black --check --line-length 120 --skip-string-normalization backend/tests/unit/test_workflow_contracts.py
• Passed lifecycle-header, version-filename, import-purity, module-isolation, workflow-contract, diff-hygiene, release-policy, and deployment-concurrency checks.
• The repository-wide Firestore query ratchet still fails on Windows with 40 unregistered and 3 unsupported fingerprints. The same command reports the identical fingerprints from an unchanged detached origin/main worktree, so that baseline is unrelated to this test-only diff.

Tests

The changed file is the regression coverage. Its existing 19 contract tests still exercise the same assertion set, now under the same strict fast-unit timing guard that exposed the Windows failure.

Root cause and durable guard (bug fixes)

Each of seven tests independently imported select_backend_unit_tests.py and walked the complete unit-test tree. Windows filesystem startup variance pushed that repeated discovery over the fast-unit per-file budget even though every assertion passed. A module-scoped fixture performs the immutable discovery once, and the existing duration guard continues to enforce the budget.

[Git-on-my-level]

Thanks for this refactor, @tianmind-studio — clean approach.

What I checked

• The module-scoped selector_and_all_tests fixture correctly replaces per-test _load_script("select_backend_unit_tests") + discover_all_tests() across all seven tests, performing the immutable filesystem discovery once instead of seven times.
• No assertion logic changed — the fixture returns (selector, all_tests) and each test unpacks it exactly as before.
• The @pytest.fixture(scope="module") scope is the right choice: discovery is deterministic and side-effect-free within a module run, so caching it is safe.

Checks
All check-runs on this head are green (Formatting, Hygiene, Detect Changes, Backend unit suite, Hermetic Backend E2E, etc.). The cancelled runs in the CI timeline are superseded re-runs, not failures.

Verdict
Good test-infrastructure improvement that directly addresses #9729. The Windows timing root cause (repeated discovery) is well explained in the PR body, and the fix is idiomatic pytest. Leaving this for a maintainer to merge.

---

by AI on behalf of David — if you need David’s attention urgently, please @Git-on-my-level and escalate with need human response.

[kodjima33]

Reviewed — looks good; not auto-merging (over scope/size or non-bug area), landing left to maintainer.

[kodjima33]

test(backend): reuse workflow selector discovery; CI green

[kodjima33]

test(backend): reuse workflow selector discovery; CI green