Make sure boxplot check doesn't polute other stats/geoms

[thomasp85]

Superseedes #3764

This PR introduces a new switch in has_flipped_aes() which confines the single value check to those requesting it (currently only boxplot)

[yutannihilation]

Ah, now I understand your comment on #3764. The example I used was the one that should not be flipped, and I confirmed this PR handles them correctly. Thanks.

df <- data.frame(value = 1:3, group = c("a", "a", "b"))

devtools::load_all("~/Documents/repo/ggplot2/")
#> Loading ggplot2

l <-  stat_summary(fun = "mean", geom = "bar", position = "dodge")
p1 <- ggplot(df, aes(value, 1, fill = group)) + l
p2 <- ggplot(df, aes(value, 0, fill = group)) + l

patchwork::wrap_plots(p1, p2, ncol = 1)

^{Created on 2020-02-06 by the reprex package (v0.3.0)}

[yutannihilation]

StatBoxplot fills 0 here, so this PR looks the right fix to me (though I still don't understand where all(x = 1) came from...).

ggplot2/R/stat-boxplot.r

Line 51 in 6207d2f

data$x <- data$x %||% 0

Can we add a simple test for this? If you think it's not worth adding, I'm fine since this seems very corner case anyway.

[thomasp85]

I have no idea how I convinced myself it defaulted to 1 🤷‍♂️