Support custom bin edges for histogram binning (nbins only accepts an int)

[tpvasconcelos]

Problem

When using histogram binning (ridgeplot(samples=..., nbins=...)), the bins are always nbins equal-width bins spanning [min(samples), max(samples)] (numpy's default). For discrete/count data this means bars are generally not aligned with the sample values. For example, for samples=[[0, 1, 2, 3]] and nbins=4, the bin edges are [0, 0.75, 1.5, 2.25, 3.0] and the bars are drawn at the true bin centers 0.375, 1.125, 1.875, 2.625.

This is correct histogram geometry (and consistent with matplotlib/seaborn), but users with discrete/count data often want integer-aligned bins, i.e. edges at -0.5, 0.5, 1.5, ... so that each bar is centered on an integer value. There is currently no way to express this: nbins only accepts an int, so the only workaround is to precompute the histogram manually and pass densities directly.

Proposal

Accept a sequence of bin edges in addition to an int, mirroring np.histogram's bins parameter:

# Today (equal-width bins over [min, max]):
ridgeplot(samples=samples, nbins=4)

# Proposed (custom edges, e.g. integer-aligned bins for count data):
ridgeplot(samples=samples, nbins=[-0.5, 0.5, 1.5, 2.5, 3.5])

(Or via a new dedicated parameter, if overloading nbins is undesirable.)

Notes

• np.histogram already supports edge sequences natively, so the change in ridgeplot._hist.bin_trace_samples should be small; most of the work is API design, typing, and docs.
• Surfaced during the test-suite review in Improve test suite: ridgeplot._hist #365.