SQLite has rather unfortunate defaults for concurrent accesses, leading to database locked errors. https://github.com/matrix-org/dendrite/pull/1290 appears to try to fix some of this, but doesn't appear to cover a concurrent read+write and the errors still happened in large quantities when I joined the dendrite room.
This PR enables WAL for concurrent read+write and much better performance. It also sets busy_timeout to 10 seconds to not immediately fail operations while the database is locked, providing a DB level solution to PR 1290's objective with additional coverage.
I chose this particular location to add the two statements as they should apply to every SQLite DB connection.
From a quick test it looks as if the db locked errors are gone with this change.
