Fix geom_ribbon() on non-cartesian Coords

[yutannihilation]

Fix #4023

The current implementation squashes the upper and lower lines together and tweaks the lower one's id on the assumption that the upper and the lower has the same number of rows as the original data. But, it's not true when the Coord is non-cartesian.

ggplot2/R/geom-ribbon.r

Lines 131 to 135 in 4826838

	positions <- new_data_frame(list(
	x = c(data$x, rev(data$x)),
	y = c(data$ymax, rev(data$ymin)),
	id = c(ids, rev(ids))
	))

ggplot2/R/geom-ribbon.r

Line 139 in 4826838

munched <- coord_munch(coord, positions, panel_params)

ggplot2/R/geom-ribbon.r

Lines 158 to 164 in 4826838

	# increment the IDs of the lower line
	munched_lines$id <- switch(outline.type,
	both = munched_lines$id + rep(c(0, max(ids, na.rm = TRUE)), each = length(ids)),
	upper = munched_lines$id + rep(c(0, NA), each = length(ids)),
	lower = munched_lines$id + rep(c(NA, 0), each = length(ids)),
	abort(glue("invalid outline.type: {outline.type}"))
	)

This PR fixes this issue by processing the upper and the lower separately.

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

d <- base::data.frame(x = c(1, 2), y = c(3, 4))

ggplot(d) +
  geom_area(aes(x, y), colour = "black", fill = "pink") +
  coord_polar()

^{Created on 2020-05-24 by the reprex package (v0.3.0)}

[yutannihilation]

@thomasp85
If there's still time left, can we include this in v3.3.1 since this is a kind of regression bug introduced in v3.3.0? If not, I think it's ok to wait for another release, though.

[clauswilke]

For what it's worth, I think it would be good to merge this and #4026 before the release.

[thomasp85]

I don’t see any problem merging both of these

[yutannihilation]

Thanks, then I'm merging this.