imstat vs astropy/python stats

[SaOgaz]

Look into differences in run time and result values. See #54 for some background.

[SaOgaz]

Run time differences is being fixed in astropy/astropy#7478.

We tracked the inconsistent result values into the difference in the clipping. Numpy uses a simple median as the center point for clipping, while IRAF does the following:

the midpoint is estimated by integrating the histogram and computing by interpolation the data value at which exactly half the pixels are below that data value and half are above it

You can define the midpoint in numpy through the cenfunc parameter for any users that really need to exactly replicate the IRAF algorithm. I will close this issue when I've updated the notebook with this note.