Fix the logic for skipping _repo_setup_updates
    
    Just checking for the package update list is wrong in a couple of
    ways:
    
    1. It goes wrong for child tests that inherit the main disk from
    the parent test, but not the second disk where the actual update
    repo lives. They'll have the package list, but not the repo. So
    we should test for both the package list and the repo dir.
    
    2. It goes wrong for side tag tests, because we don't write a
    package list (or create an update repo) on that path. So on that
    path let's check whether we already created the repo definition
    we use on that path.
    
    Signed-off-by: Adam Williamson <awilliam@redhat.com>