test(user-settings): cover Mongo unparseable-timestamp read path

[mauripunzueta]

Summary

Adds one MongoDB SettingsStore test that mirrors SQLite's existing get_settings_tolerates_unparseable_timestamp coverage: it plants a row with a valid JSON document but a non-datetime updated_at, then asserts get_settings still returns the document and version — exercising the timestamp-fallback branch (get_datetime(...).map(...).unwrap_or_else(|_| Utc::now())) instead of failing the read.

This closes the last cleanly, deterministically testable gap in mongodb/user_settings.rs, bringing its error-path suite to parity with the SQLite backend. The remaining uncovered lines there are DB-error-injection .map_err arms and concurrency-race branches — the same class the PostgreSQL backend also leaves uncovered — which aren't worth fault-injection or flaky timing tests.

Stacking

Base branch is feat/user-ui-settings (PR #151), not main, because the MongoDB SettingsStore this test targets is introduced by #151 and does not exist on main.

Once #151 merges, this PR will be retargeted to main (and rebased), at which point it picks up full CI. Until then, as a stacked PR (base ≠ main) it does not trigger the HFS CI workflow.

Test plan

• cargo test -p helios-persistence --features mongodb --test mongodb_tests mongodb_integration_settings_get_tolerates_unparseable_timestamp (runs against a Mongo testcontainer, or HFS_TEST_MONGODB_URL if set; skips gracefully when neither is available).

[mauripunzueta]

Closing — wrong scope. This is a narrow, one-off test of the MongoDB SettingsStore (which only exists in #151), not the reusable, broadly-useful MongoDB test infrastructure that was actually requested. Superseding with a test that branches from main and helps any future code touching the MongoDB backend. Branch kept for now in case we want the parity test inside #151 later.