[Checkers] Nullness inference

[Michael Ernst]

Mahmood-

Your generalization of my suggestion seems correct; you're right that I
didn't spell out all the details, but that's essentially what I had in
mind.  Thanks!

Technically, I like this increased precision.  The question that remains
for me is how important this is.  Fausto has already shown that it can come
up in practice, so that isn't the question.  I wonder if a programmer would
prefer his more precise (and more tricky) type, or if there are places that
it enables other precision that eases type-checking or understanding.  I
could go either way -- I just don't have any evidence.

                    -Mike