MB has found restricting pystare conversion calls to numpy arrays is a win.

[michaelleerilee]

We might wish to add routines to the API that feature numpy variables to avoid slow-downs apparently caused by Python's type boxing.

I don't know if we would want to do this via a namespace as in

pystare.numpy.from_latlon(...)

or via a function names, as in

pystare.from_latlon_np(...)

We also might want to treat the issue of masked arrays and fill-values (#106) at the same time with the numpy-only signatures.