fix: harden kanban sqlite writes

[Oceanswave]

Summary

• add an explicit PRAGMA busy_timeout=30000 to every Kanban DB connection
• annotate Kanban SQLite connections with the resolved board DB path
• serialize same-process write transactions through a per-board Python RLock before BEGIN IMMEDIATE
• keep the existing SQLite WAL/rollback-journal fallback as the cross-process authority

Tests

• python -m pytest tests/hermes_cli/test_kanban_db.py::test_connect_sets_explicit_busy_timeout tests/hermes_cli/test_kanban_db.py::test_write_txn_uses_per_database_python_lock -q
• python -m pytest tests/hermes_cli/test_kanban_db.py tests/hermes_cli/test_kanban_notify.py -q

Context

Follow-up hardening after a real Kanban board DB corruption/recovery incident. This does not replace SQLite's locking; it prevents local gateway/notifier/CLI threads from stampeding the same board DB and makes the cross-process busy wait explicit.

Fixes #32532
Refs #31158

[Oceanswave]

Closing because the upstream database hardening changes have now landed, so this duplicate Kanban SQLite write-hardening PR is no longer needed.