Use busy_timeout for better lock handling across Isolates

[rkistner]

While read statements can execute concurrently on multiple connections, write statements and transactions use an exclusive lock. Currently, a global database-level lock is maintained and exposed across isolates, but this requires explicitly passing an IsolateConnectionFactory across Isolates. This is fine when explicitly spawning a new Isolate to do background work from the main Isolate, but does not work when the Isolates are managed externally, for example when using flutter_workmanager.

This configures SQLite busy_timeout to 30 seconds to work around the issue. When another connection is locking the database, SQLite will retry until the timeout has been reached.

In principle that makes our own global mutex redundant, and removing it can give some performance gains. That will form a separate PR - see #35.