Switch to using numpy to calculate bin edges in histogram where possible and calculate bin edges in consistent way #8011

numpy now has several methods for estimating bin widths. Some are the same as those in astropy, some are different. When possible, we should use the numpy methods (e.g. for 'scott' and 'freedman' (which numpy calls 'fd')).

There is also a difference between the way astropy calculates the bins for these cases and the way numpy does. In astropy, the bin width calculated from the estimator is used without modification, with the result that the last bin edge is often larger than the range or the last data point (see, e.g. https://github.com/astropy/astropy/blob/master/astropy/stats/histogram.py#L146) . In numpy, the upper bin edge is always either the last data point or the range; that results in a modestly smaller bin size than you get from the estimator (their calculation is at https://github.com/numpy/numpy/blob/v1.15.1/numpy/lib/histograms.py#L375).

I don't know that the numpy way is superior but since we are trying to offer a drop-in replacement I think we should follow their convention.

Sorry for the odd combination of labels -- I think we also ought to use numpy to make available the estimators that numpy has that astropy doesn't (numpy's list is at: https://docs.scipy.org/doc/numpy-1.15.1/reference/generated/numpy.histogram_bin_edges.html#numpy.histogram_bin_edges).

So the bug is our inconsistency regarding bin edges, the enhancement is adding new estimators.