NSCell mouse tracking rewrite

[Fred Kiefer]

I just pushed a commit where the whole mouse tracking logic in NSCell
gets changed. This will of course affect all the subclasses of NSCell as
well. I am convinced that the new code is better than the old one, but
there may be other code that relies on the old behaviour and now needs
adjustment.
I tested a lot of these subclasses, but there are way to many
combinations for me to check. Could you all please see whether the
behaviour in any of our favourite GNUstep application is still
consistent and report back any changes?

Thank you
Fred