fix(config,startup): accept and wire session_state_path (#653)

[Yatsuiii]

Closes #653. Credit to @saintmalik for the report - the root-cause tracing to both defects, including the exact config.py/startup.py line citations, is theirs; this PR is the fix.

session_state_path was documented (docs/configuration.md) and its feature described as shipped (CHANGELOG, merged #629), but was non-functional in two independent ways:

1. _KNOWN_TOP_KEYS (config.py) omitted the key, so load_config rejected any config that set it, before the existing parsing code (config.py:489-491, :561) ever ran.
2. Even past that, run_startup never constructed a SqliteSessionStateStore - only an import and a RuntimeContext dataclass field existed. RuntimeContext.session_state_store stayed None regardless of configuration, so the shared, persistent session-sensitivity ratchet never activated: a restart still lost the accumulated value, and multiple instances serving one session still ratcheted independently.

Confirmed both by direct inspection before writing anything: SqliteSessionStateStore( had zero call sites in src/ outside its own class definition.

What changed

• config.py: added session_state_path to _KNOWN_TOP_KEYS.
• startup.py: added _open_session_state_store(), called from run_startup as step 5f. Opens the store when session_state_path is configured and wires it into RuntimeContext; leaves it None, unchanged, when unconfigured. An unopenable path fails startup closed (SESSION_STATE_STORE_UNAVAILABLE), mirroring the existing audit_store open-failure pattern exactly.
• Extracted into its own helper rather than inlined, to avoid adding to run_startup's already-high pre-existing complexity (measured at 26 on clean main before this change; unchanged after, since the helper is a single call site).

Scope

The store's own logic (locking, hydrate/reset semantics, cross-process behavior) is untouched - tests/unit/test_session_state_store.py's existing suite already exercises SessionState.hydrate() against a real SqliteSessionStateStore, including a real multiprocessing-based cross-process lock test. This PR only closes the wiring gap between that already-tested mechanism and run_startup.

Testing

• New: config acceptance test, and three run_startup tests (unconfigured stays None; configured opens a real store and the file exists on disk, closed in finally; a store-open failure aborts startup - forced via monkeypatch rather than an OS-level directory trick, since this repo's CI runs both Linux and Windows).
• Full suite: pytest tests/unit -q -> 1838 passed, 6 skipped, one pre-existing unrelated failure (test_startup_fails_on_unknown_tee_provider_name, fails identically on clean main, a local TPM auto-detection quirk unrelated to this change).
• Ruff and mypy clean on all touched source files.

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!