[BUG] `hist.numpy.histogram` from array with NaN entries

[alexander-held]

Describe the bug

When giving an array of values to hist.numpy.histogram that contains NaN values, the histogram creation fails:

>>> import hist
>>> import numpy as np
>>> hist.numpy.histogram([1, 2, np.nan])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "[...]/lib/python3.9/site-packages/boost_histogram/numpy.py", line 166, in histogram
    result = histogramdd(
  File "[...]/lib/python3.9/site-packages/boost_histogram/numpy.py", line 78, in histogramdd
    cpp_ax = _core.axis.regular_numpy(b, r[0], r[1])
ValueError: forward transform of start or stop invalid

I assume that the challenge is the automatic detection of a suitable bin range, since this does not work with pure numpy either:

>>> np.histogram([1, 2, np.nan])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<__array_function__ internals>", line 180, in histogram
  File "[...]/lib/python3.9/site-packages/numpy/lib/histograms.py", line 793, in histogram
    bin_edges, uniform_bins = _get_bin_edges(a, bins, range, weights)
  File "[...]/lib/python3.9/site-packages/numpy/lib/histograms.py", line 426, in _get_bin_edges
    first_edge, last_edge = _get_outer_edges(a, range)
  File "[...]/lib/python3.9/site-packages/numpy/lib/histograms.py", line 323, in _get_outer_edges
    raise ValueError(
ValueError: autodetected range of [nan, nan] is not finite

Perhaps this is more of a feature request than a bug, but I think it would be useful to automatically mask out NaN values so that histogramming could still work (perhaps with an accompanying warning), or otherwise to catch this with a more descriptive error message.

This came up via uproot-browser, and it is the issue I ran into that prompted scikit-hep/uproot-browser#29.

Steps to reproduce

see above, Python 3.9.10, relevant library versions:

boost-histogram               1.3.1
hist                          2.6.1
numpy                         1.22.2