Comment 14 for bug 135956

> Johannes - well, .cue files specify the file names they are associated
> with in themselves, so as long as we parse any .cue files in a directory
> first so we know which 'real' files to ignore, there shouldn't be any
> need to assume.

I'm more thinking about the case if the cue file is referring to a
file in a different directory (the spec allows this), in which case
the file may have been scanned previously.

Obviously we can still handle this perfectly, by removing a file from
the collection if we find a cue pointing there. Just a bit more work.