Support boost_histogram/hist's categorical axes

[jpivarski]

Inspired by #659 (@jlabounty). Currently, the boost-histogram/hist → ROOT TH* conversion takes a categorical axis with n categories as a continuous regular axis of n bins from 0 to n-1 (inclusive). ROOT TH* objects have no notion of categorical axes, but they do have fLabels, which can be set to a list of strings. When this value is set to n strings, ROOT draws these strings under each tick, effectively simulating a categorical axis.

This feature request would be (1) to have boost-histogram/hist → ROOT TH* set fLabels, somewhere here (may have to be expanded to a regular for loop):

uproot5/src/uproot/writing/identify.py

Lines 320 to 331 in c7d7282

	# convert all axes in one list comprehension
	axes = [
	to_TAxis(
	fName=getattr(axis, "name", default_name),
	fTitle=getattr(axis, "label", getattr(obj, "name", "")),
	fNbins=len(axis),
	fXmin=axis.edges[0],
	fXmax=axis.edges[-1],
	fXbins=_fXbins_maybe_regular(axis, boost_histogram),
	)
	for axis, default_name in zip(obj.axes, ["xaxis", "yaxis", "zaxis"])
	]

And (2) to interpret a TAxis with fLabels is not None and len(fLabels) != 0 as a boost-histogram/hist categorical axis when transforming the other way:

uproot5/src/uproot/behaviors/TH1.py

Lines 17 to 40 in c7d7282

	def _boost_axis(axis, metadata):
	boost_histogram = uproot.extras.boost_histogram()
	fNbins = axis.member("fNbins")
	fXbins = axis.member("fXbins", none_if_missing=True)
	if axis.member("fLabels") is not None:
	out = boost_histogram.axis.StrCategory([str(x) for x in axis.member("fLabels")])
	elif fXbins is None or len(fXbins) != fNbins + 1:
	out = boost_histogram.axis.Regular(
	fNbins,
	axis.member("fXmin"),
	axis.member("fXmax"),
	underflow=True,
	overflow=True,
	)
	else:
	out = boost_histogram.axis.Variable(fXbins, underflow=True, overflow=True)
	for k, v in metadata.items():
	setattr(out, k, axis.member(v))
	return out

We probably won't be able to preserve the int-ness of IntCategorical (fLabels are strictly strings), but maybe if all strings don't raise ValueError when cast as a Python int, that would be a good indicator.