fix stat_contour() for irregular data grids.

[clauswilke]

Closes #3906.

library(ggplot2)

v <- ggplot(faithfuld, aes(waiting, eruptions, z = density)) +
  geom_point(aes(color = density)) + geom_contour(color = "white")

v

v + scale_x_log10()

^{Created on 2020-03-21 by the reprex package (v0.3.0)}

[paleolimbot]

Just a few comments for your consideration!

[paleolimbot]

This looks great! I wrote this based on the code in GeomRaster because it was faster...before it was:

raster <- tapply(data$z, data[c("x", "y")], identity)

(I found that here in case you want to look at the context)

It might be worth checking whether that is faster (I suspect your way is better)

[clauswilke]

tapply() is ~40% slower.

As an aside, we're doing a lot of work wrangling the data into the right format for isoband and then back out. It'd probably be more efficient to just write an isoband function that takes the data in the format preferred by ggplot2 and returns it in a format that can be used directly. But that's for another day.