elliptical_overlap_grid generating nan's in some cases

[tepickering]

When setting up regression tests for elliptical_overlap_grid, I ran into a number of failures where the masks were filled with nan. The ones I tracked down all had equal semi-major and semi-minor axis sizes. An example failure case is:

In [4]:
m = elliptical_overlap_grid(-1, 1, -1, 1, 1000, 1000, .8, .8, 0.5, 1, 10)
m.mean()
Out[4]:
nan

Slightly changing one of the axis values yields appropriate numbers:

In [13]:
m = elliptical_overlap_grid(-1, 1, -1, 1, 1000, 1000, .8, .80001, 0.5, 1, 10)
m.max()
Out[13]:
1.0000000000001465

Setting use_exact to 0 also yields correct results in all cases so the bug is somewhere within the use_exact code.

[bsipocz]

Closing as this was fixed in #276.