Do not use class information for raster resolution calculation

[teunbrand]

This PR to the RC aims to fix a bug uncovered during reverse dependency checks.

When feeding continuous data to geom_raster() but using discrete scales confers the mapped_discrete class even upon the continuous data. We use the heuristic that mapped_discrete values have resolution 1 in #5247. This leads to problems in geom_raster():

library(ggplot2)

t <- c(0.1, 0.9, length.out = 2)
df <- expand.grid(c(0.1, 0.9), c(0.1, 0.9))
df$fill <- LETTERS[1:4]

# Should show 4 squares
ggplot(df, aes(Var1, Var2, fill = fill)) +
  geom_raster() +
  scale_x_discrete() +
  scale_y_discrete()

In this PR, the data of geom_raster() is unclass()'ed so that resolution correctly pick up on its continuous nature:

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

ggplot(df, aes(Var1, Var2, fill = fill)) +
  geom_raster() +
  scale_x_discrete() +
  scale_y_discrete()

I found out about this using case via the breakage of stars, where the following example would consume gigabytes of memory due to building an enormous raster. With this PR, this is now more in sync with the size of the original data.

library(stars)

x <- system.file("tif/L7_ETMs.tif", package = "stars") |>
  read_stars()

ggplot() + 
  geom_stars(data = x) +
  coord_equal() +
  facet_wrap(~band) +
  theme_void() +
  scale_fill_viridis_c() +
  scale_x_discrete(expand = c(0, 0)) +
  scale_y_discrete(expand = c(0, 0))

^{Created on 2024-01-12 with reprex v2.0.2}

[thomasp85]

LGTM