BUG : fix non-uniform grids in pcolorfast

[tacaswell]

Closes #4227

attn @mdboom @efiring This fixes the problem, but I have no idea why (I just looked at the c++ header, it wanted ints so I made sure it get ints).

[tacaswell]

We might also want to back-port this to color-overhaul.

[efiring]

Looks good to me. The only question it raises in my mind is whether after multiplication by magnification, a rounding operation would be better than truncation.

[efiring]

There is no point in backporting because the CXX version was handling the conversion to int. That also tells me there is no need to add a rounding step; what you have is fine.
Another way to fix the problem would be to put the conversion to int back in the C++; that makes the _pcolor2 function friendlier, but given that it is private, I have no objection to letting it be fussy about inputs. It's simpler.

[tacaswell]

@efiring rebased to fix conflict in tests.

Re-reading this I read your suggestion about rounding, implemented it and then re-read your comment about not implementing it.

I think the rounding is ok as this is determining how many pixels in the agg layer that the image will take up. Assuming travis passes, I think we should leave it as-is.

[tacaswell]

well, of course travis is going to pass, we have no test coverage on this...

[efiring]

Needs a rebase? Otherwise, it looks fine.

[tacaswell]

@efiring rebased.