Now it can be anything.

for the factor_value_disagreement() values since the code actually supports this.

to make it robust to very large images.

But nevertheless, it's fixed.

infinite weights on some edges.

to static_casts whenever it was possilbe to do so.

The manpages seemed to indicate that the previous code was
correct so I'm not sure if this was a bug or not.  It's been
in dlib for years without causing trouble though.

 - Added an add() and subtract() that works on
   sparse and dense vectors.