Fixes #202.

• Adds cleanup_databases=False argument. Default False to retain original behavior
• Cleans up individual test databases during the natural fixture teardown phase, fairly straightforward
• Cleaning up the template databases is a lot more involved and a majority of the refactor being done by this PR
  • Internally we now create 1 new (randomly named) session fixture per create_postgres_fixture call
  • that gets hackily injected into the calling code's global scope under a uniqified name
  • we then use the resultant fixture to provide us a scope from which we can perform tear-down logic to delete the template databases at exit.

It probably makes sense to follow this up with a pmr --cleanup or the like kind of CLI extension, that more directly just deletes all pmr_* named databases. If there's a hard test failure, you enter a debugger and exit uncleanly, you ctrl+c enough, or various other means: it's still fairly straightforward to end up with left over databases.

1 unresolved issue:

• pytest-asyncio has all sorts of weird stuff going on in both prior versions that people claim works, and the new versions are just broken apparently. In either case, I wasn't able to find a way that seemed to work reliably in async code (at least, using pytest-asyncio). Only affecting template database cleanup.

  So unless/until there's a different plugin, or they resolve their stuff and we start requiring a much more recent minimum version bound for it after they do, template database cleanup will mean that async code wont share templates (and thus will be slower). Could be worse, but isn't ideal.