Put deadlock detector into its own background worker

We do some transactions in the maintenance daemon which can get blocked on a lock. In that case the deadlock detector no longer runs. It would be good to put the deadlock detector in its own background worker such that it never blocks and could even detect deadlocks that involve the maintenance daemon itself.

Start by locating the maintenance daemon and deadlock detector entry points in the C extension; the issue names no files or tests. Map how lock waits currently affect detector execution, then define done as detection continuing independently and covering deadlocks involving the maintenance daemon.

Written by the indexing model from the issue text.