Stabilize delete-account tests stubs on Windows

[tianmind-studio]

Summary

• Include pytz in the auto-stubbed import prefixes for both delete-account tests that import routers.users.
• Clear any existing modules under the stubbed prefixes before installing each test meta-path finder, so stale sys.modules entries from earlier tests cannot bypass the local _StubFinder.

Why

test_delete_account_purge_storage.py and test_delete_account_stripe_cancel.py both import routers.users, which has a broad import graph. In a lightweight Windows test environment, collection failed first on missing pytz. In broader collect-only runs, earlier tests can also leave stale database.* or utils.* entries in sys.modules; because importlib returns existing modules before consulting these tests' finders, stale modules can break the router import before the deletion assertions run.

This keeps both delete-account import sandboxes deterministic and scoped to the namespaces they already declare as stubbed.

Testing

• python -m pytest tests\unit\test_delete_account_purge_storage.py -q -> 5 passed
• python -m pytest tests\unit\test_delete_account_stripe_cancel.py -q -> 3 passed
• python -m pytest tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py -q -> 8 passed
• python -m pytest tests\unit\test_byok_security.py tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py --collect-only -q -x -> 89 tests collected
• python -m black --line-length 120 --skip-string-normalization tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py --check
• python -m py_compile tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py
• git diff --check origin/main...HEAD

[greptile-apps]

Greptile Summary

This PR makes two targeted changes to test_delete_account_purge_storage.py to fix flaky collection on Windows minimal venvs: it adds 'pytz' to the auto-stub prefix list so importing routers.users no longer requires the optional package, and it sweeps sys.modules clean of every stubbed namespace before installing the _StubFinder, preventing stale mocks left by earlier test files from bypassing the finder.

• 'pytz' is appended to _STUB_PREFIXES, eliminating the ModuleNotFoundError that surfaced on Windows when the minimal venv did not include the optional timezone library.
• A new module-level helper _remove_module_tree is called for every stub prefix before the _StubFinder is inserted into sys.meta_path, making the import sandbox deterministic regardless of test-collection order.
• The identical sister file test_delete_account_stripe_cancel.py uses the same _StubFinder / _STUB_PREFIXES pattern to import routers.users and is not updated, leaving the same Windows pytz gap and stale-modules vulnerability unfixed there.

Confidence Score: 4/5

Safe to merge for purge_storage tests; however the same Windows pytz failure and stale-mock bypass remain unfixed in the sister file test_delete_account_stripe_cancel.py.

The two changes to test_delete_account_purge_storage.py are correct and well-scoped. The gap is that the identical import-sandbox pattern in test_delete_account_stripe_cancel.py does not receive the same pytz stub entry or the pre-install sys.modules sweep, leaving it vulnerable to the same Windows collection failure and stale-module bypass that motivated this PR.

backend/tests/unit/test_delete_account_stripe_cancel.py — identical stub pattern without the pytz entry and without the _remove_module_tree sweep.

Important Files Changed

Filename	Overview
backend/tests/unit/test_delete_account_purge_storage.py	Adds pytz to the stub prefix list and introduces a pre-install sys.modules sweep to clear stale mocks left by earlier test files; logic is correct for this file, but the identical sister file test_delete_account_stripe_cancel.py does not receive the same fixes.

Comments Outside Diff (1)

1. backend/tests/unit/test_delete_account_purge_storage.py, line 38-53 (link)

   Sister file test_delete_account_stripe_cancel.py has the same unfixed gaps

   test_delete_account_stripe_cancel.py is a near-identical copy of this file: it uses the same _StubFinder / _STUB_PREFIXES pattern, it imports routers.users, and that module's import chain pulls in pytz. On a minimal Windows venv where pytz is absent, running stripe_cancel in isolation hits the same collection failure that this PR fixes here. That file also lacks the _remove_module_tree sweep, so any stale database.* or utils.* mock left by test_byok_security.py can bypass its _StubFinder for the same reason described in the PR.

   The two fixes applied here ('pytz' in _STUB_PREFIXES and the pre-install sys.modules clear) should be mirrored in test_delete_account_stripe_cancel.py.

_{Reviews (1): Last reviewed commit: "test(backend): isolate delete account pu..." | Re-trigger Greptile}

[tianmind-studio]

The Greptile P1 summary was based on the earlier commit that only updated test_delete_account_purge_storage.py.

Addressed in the current head 534ad233: the same pytz stub entry and _remove_module_tree(...) pre-install cleanup are now mirrored in test_delete_account_stripe_cancel.py as well. I rechecked the PR diff and both delete-account test files carry the same import-sandbox fix.

Validation from the PR body:

• python -m pytest tests\unit\test_delete_account_purge_storage.py -q -> 5 passed
• python -m pytest tests\unit\test_delete_account_stripe_cancel.py -q -> 3 passed
• python -m pytest tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py -q -> 8 passed
• python -m pytest tests\unit\test_byok_security.py tests\unit\test_delete_account_purge_storage.py tests\unit\test_delete_account_stripe_cancel.py --collect-only -q -x -> 89 tests collected

[kodjima33]

Test-only stabilization (backend/tests/unit), CI green