Fix BoundaryNorm interpolation with numpy 1.7rc.

[efiring]

Closes github issue #1258.

[WeatherGod]

I can confirm that this fixes the issue. Admittedly, I guess I would rather see the casting rules be a bit smarter in numpy when performing in-place operations. Because of this, we lose the ability to do an in-place multiplication, which could be a performance hit when the Norm is applied to much larger datasets.

[pelson]

@WeatherGod : I agree with your concerns, but don't feel that it should hold us up in getting a RC2 ready. Are you willing to merge & open a seperate issue (if there is something that can be done by us, and not in numpy itself)?

[WeatherGod]

Heh, just noticed that the relevant code was actually recently changed to do the inplace operations (perhaps in one of the PEP8 changes?). Before, it was a single line doing the multiplication and casting and assigning back to the iret array. I still prefer if the in-place multiplication would work, but I am going to raise that issue on the numpy list instead. I will go ahead and merge this in.

[efiring]

Yes, it was in the PEP8 change, #1a07db586e.