Treat binwidth as approximate to avoid dropping outermost datapoints

[mwaskom]

This PR changes the interpretation of the binwidth parameter in histplot and objects.Hist.

With the previous implementation, floating point errors could cause the largest datapoint(s) to be silently dropped. The solution here is to always honor the bin range (either the explicit as specified through binrange or implicit as computed from the data range) and to make the actual bin width only approximately equal to the binwidth parameter when binwidth does not evenly divide it.

I could not think of a simple + robust solution to detect and handle floating point issues while also reliably honoring expected bin ranges, and I think that the expected violation of expectations will be smaller this way, although it is a minor API change. In most cases it should not be noticeable.

Fixes #3220

[codecov]

Codecov Report

Merging #3489 (ba0f186) into master (c01e103) will decrease coverage by 0.01%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3489      +/-   ##
==========================================
- Coverage   98.33%   98.33%   -0.01%     
==========================================
  Files          75       75              
  Lines       24570    24569       -1     
==========================================
- Hits        24162    24161       -1     
  Misses        408      408              

Files Changed	Coverage Δ
seaborn/_stats/counting.py	100.00% <100.00%> (ø)
tests/test_distributions.py	99.80% <100.00%> (ø)