Add Bars for more efficient bar plots and improve Bar as well

[mwaskom]

Similar to how we have Line/Lines, the Bars mark uses a PatchCollection for more efficient drawing. (Unlike with Lines, we don't need to reduce the mappable properties). This can have a substantial effect:

x = np.linspace(-3, 3, 1000)
y = np.exp(-(x ** 2))

%time so.Plot(x, y, color=x).add(so.Bar()).save(io.BytesIO())
%time so.Plot(x, y, color=x).add(so.Bars()).save(io.BytesIO())

CPU times: user 2.66 s, sys: 20.3 ms, total: 2.68 s
Wall time: 1.65 s
CPU times: user 918 ms, sys: 12.3 ms, total: 930 ms
Wall time: 397 ms

The default property values for Bars are also a little different: width is set to 1, edges use the patch.linecolor rc parameter, and edge widths are automatically set using the same approach in histplot to scale down for a dense plot.

so.Plot(diamonds, "price").add(so.Bars(), so.Hist())

[codecov]

Codecov Report

Merging #2893 (6415872) into master (3180fd7) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #2893      +/-   ##
==========================================
+ Coverage   98.19%   98.20%   +0.01%     
==========================================
  Files          69       69              
  Lines       22435    22561     +126     
==========================================
+ Hits        22030    22156     +126     
  Misses        405      405              

Impacted Files	Coverage Δ
seaborn/_marks/bars.py	100.00% <100.00%> (ø)
seaborn/_marks/base.py	98.30% <100.00%> (+0.04%)	⬆️
seaborn/_marks/scatter.py	100.00% <100.00%> (ø)
seaborn/objects.py	100.00% <100.00%> (ø)
tests/_marks/test_bars.py	100.00% <100.00%> (ø)
tests/_marks/test_base.py	100.00% <100.00%> (ø)