Add an installed qualification command for temporal memory

[dcellison]

Third of three deliveries for #1717.

Change

python -m kai.workshop.memory_qualification --db <DATA_DIR>/kai.db --snapshot <file> qualifies temporal memory on an installed Kai. It opens the database read-only and never the vector store, so it runs while the service is up. It passes when:

1. every memory diagnostic line install status prints is active (current truth, reconciliation, episode history, extraction receipts), which means their integrity, replay, projection, and drift gaps are all zero;
2. every owner the service counted has a stored vector audit, and the oldest audit is newer than the latest projection (otherwise it asks for a restart or a search index check);
3. conversational backend and model settings are unchanged: the first run records them in the snapshot file, later runs compare against it.

Output is one PASS or FAIL line per check, with no memory content. Exit status is 0 on pass and 1 otherwise.

The stored vector audit time is now recorded at full precision, so an audit taken in the same second as a projection is not mistaken for an older one.

Verification

• Tests: a consistent install passes and records defaults once; a changed default, a stale audit, a missing audit, and a failed projection each fail the right check; a missing database is reported.
• Run read-only against the production database: all six checks pass, and a second run compares 11 settings unchanged.
• Full suite: 7117 passed. Pyright clean.