BUG: IntNA division can alter array inplace

[jbrockmendel]

IntegerArray._maybe_mask_result starts with:

if is_float_dtype(result):
     mask |= (result == np.inf) | (result == -np.inf)

but in many cases, mask will be self._mask, so it gets altered inplace. The intuitive solution is to add a check for mask is self._mask and make a copy, but that turns out to break 32 tests, so somewhere we're relying on this incorrect behavior.

cc @jreback

[jreback]

hmm this looks buggy then as we shouldn’t alter inplace (not sure if this shows up in the wild)

[ruddyscent]

I'm working on it at PyCon Korea 2019 Sprint.