Summary

Harden Kanban handling for corrupt per-board SQLite databases.

Before this change, repeated access to the same corrupt board DB could create repeated kanban.db.corrupt.* backup files. This could grow unbounded across retries/restarts if a dashboard or dispatcher kept touching the same malformed board.

This patch makes corrupt-board handling idempotent and bounded.

Changes

• Add durable corrupt-board quarantine state via kanban.db.corrupt-quarantine.json.
• Fingerprint corrupt DB files so repeated access to the same corrupt DB does not create new backups.
• Add CORRUPT_DB_BACKUP_RETENTION = 3.
• Prune old .bak, .bak-wal, and .bak-shm corrupt backup files.
• Clear stale quarantine markers after a healthy DB open/recovery.
• Extend dashboard/API handling to return clean unreadable-board diagnostics.
• Preserve dashboard fallback away from stale hermes.kanban.selectedBoard.
• Extend gateway dispatcher handling for quarantined corrupt boards.
• Add regression coverage for idempotent quarantine, retention, dashboard diagnostics, and dispatcher handling.

Validation

venv/bin/python -m pytest tests/plugins/test_kanban_dashboard_plugin.py
97 passed, 1 warning

venv/bin/python -m pytest tests/hermes_cli/test_kanban_db.py -k "corrupt_db_quarantine or prune_corrupt_db_backups or init_db_refuses_corrupt_existing_file or connect_refuses_corrupt_existing_file or locked_healthy_db_does_not_classify_as_corrupt"
6 passed, 169 deselected

venv/bin/python -m pytest tests/hermes_cli/test_kanban_core_functionality.py -k "gateway_dispatcher_disables_corrupt_board_without_traceback or gateway_dispatcher_disables_quarantined_corrupt_board_without_traceback"
2 passed, 164 deselected