re commit: 7641959c5c31636cb9f492e76d0f0a3c8b21a82e (don't use file contents for mimetype detection)

[David Nolden]

I've committed this now. I'm not entirely happy with it, but I don't
see another way atm, because in the moment where we need the
content-based lookup most, we will disable it. However, we cannot
afford to look up the contents of thousands of files to determine the
type, because it all happens synchronously within the foreground
thread. The best solution would be if we could somehow make this check
asynchronous (could probably be done using some kioslave or similar),
but then languagesForUrl(..) couldn't work the way it does (directly
returning a result).