Improve histogram efficiency

[mwaskom]

Closes #2550

Now we only pass an array of bins into the numpy function when that's what the user supplied. Otherwise we keep track of the number of bins and bin range. This gives a ~order of magnitude performance boost for medium-large arrays (note log scales):

bins = 50
ns = np.logspace(2, 7, num=6, dtype=int)
times = pd.Series(index=ns, dtype=float)

for n in ns:
    hist = sns._statistics.Histogram(bins=bins)
    x = np.random.normal(0, 1, size=n)
    res = %timeit -o hist(x)
    times[n] = res.average

No change to external API.

[codecov]

Codecov Report

Merging #2570 (0c2548e) into master (ec8fcc9) will increase coverage by 0.00%.
The diff coverage is 100.00%.

❗ Current head 0c2548e differs from pull request most recent head b29166f. Consider uploading reports for the commit b29166f to get more accurate results

@@           Coverage Diff           @@
##           master    #2570   +/-   ##
=======================================
  Coverage   97.45%   97.45%           
=======================================
  Files          17       17           
  Lines        6332     6337    +5     
=======================================
+ Hits         6171     6176    +5     
  Misses        161      161           

Impacted Files	Coverage Δ
seaborn/_statistics.py	100.00% <100.00%> (ø)
seaborn/distributions.py	96.37% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ec8fcc9...b29166f. Read the comment docs.