NaN treatment

[jamienoss]

Dedupe NaN conditionals

src

...
if preserve_nan:
        badvals = np.isnan(array_internal)

if nan_treatment == 'fill':
        initially_nan = np.isnan(array_internal)
        array_internal[initially_nan] = fill_value
...

To prevent possible duplicated work when preserve_nan == True and nan_treatment == 'fill' the condtions should be joined:

...
if preserve_nan or nan_treatment == 'fill':
    initially_nan = np.isnan(array_internal)
    if nan_treatment == 'fill':
        array_internal[initially_nan] = fill_value
....

nan_treatment == 'fill'

For nan_treatment == 'fill' in convolve(), since any and all possibel NaN values are removed, the subsequent npy_isnan(val), for example for boundary=None (the default), in the Cython routine boundary_none.pyx can be removed. Removing this is a step closer to removing all conditionals such that vectorization can take place.