While this means we need an additional query for initial lookup, it makes the *filtering* query (which is the heavy part) require far fewer joins, thus making it more efficient.
Also adds a new unit test to verify that our filter filters to the correct set of repositories.