Add box mark

[mwaskom]

This is a rectangular mark drawn between min/max values, sort of a cross between Bar and Range:

(
    so.Plot(tips, "day", "total_bill", color="sex")
    .add(so.Box(), so.Dodge())
)

I think it's very likely that there will be a standalone mark for drawing a full box-and-whisker plot in one layer — name and implementation strategy TBD — but until then this mark provides what we need to have an (outlier-less) "box plot":

(
    so.Plot(tips, "day", "total_bill", color="sex")
    .add(so.Box(alpha=.35), so.Perc([25, 75]), so.Dodge())
    .add(so.Range(), so.Perc([0, 25]), so.Dodge())
    .add(so.Range(), so.Perc([75, 100]), so.Dodge())
    .add(so.Dash(), so.Perc([50]), so.Dodge())
)

Couple notes:

• Took a long time to decide on the name, considering other options such as Span, Rect, Plank, Beam, Lath. The first two are more useful for other planned marks, and the latter 3 are not obvious. This does commit us to using Boxplot or WhiskerBox or something like that for the proper "box plot" mark.
• This uses the edge-clipping trick that Bar and Area marks use. I'm souring on this a bit and may revert it in a future release.

[codecov]

Codecov Report

Merging #3127 (76a5abb) into master (d4d27ad) will increase coverage by 0.00%.
The diff coverage is 100.00%.

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #3127    +/-   ##
========================================
  Coverage   98.41%   98.42%            
========================================
  Files          76       76            
  Lines       24076    24190   +114     
========================================
+ Hits        23695    23809   +114     
  Misses        381      381            

Impacted Files	Coverage Δ
seaborn/_marks/area.py	96.20% <100.00%> (-0.31%)	⬇️
seaborn/_marks/bar.py	100.00% <100.00%> (ø)
seaborn/_marks/base.py	98.51% <100.00%> (+0.11%)	⬆️
seaborn/objects.py	100.00% <100.00%> (ø)
tests/_marks/test_bar.py	100.00% <100.00%> (ø)

[jcmkk3]

Observable Plot does this with the regular bar mark, but they have x1 x2 and y1 y2 position options. Vega-lite also has x x1 and y y1 position options.

That doesn't mean that Seaborn Objects should do the same, but it is interesting to see how other grammar of graphics libraries handle it.

[mwaskom]

The main practical difference would be the loss of the range calculation when xmin/xmax aren't specified; so that first example would become:

(
    so.Plot(tips, "day", "total_bill", color="sex")
    .add(so.Bar(), so.Perc([25, 75]), so.Dodge())
)

I do wonder a little bit whether the Box mark should have a median line when it computes a range (but should it have one when there are only two values, as in the Perc example? That's probably too hard to explain / predict).

One thing that would be a little awkward to do with this parameterization would be a candlestick chart when your data naturally has "open" and "close" columns. Having a y/y1 or y1/y2 parameterization is a little less opinionated there compared to ymin/ymax (although actually, given the way the mark is implemented, you sill get the expected result:

so.Plot(["a", "b"], ymin=[1, 2], ymax=[2, -1]).add(so.Box())