annotation_raster()/_custom() respond to scale transformations

[teunbrand]

This PR aims to fix #3120 and is incompatible with #3121.

Briefly, it uses the panel parameters to transform the corners of the rectangle used for annotation placement.

To recap some discussion in #3121. The scales should affect the annotation, but not vice versa (i.e., we don't want to train scales on annotations) #3121 (comment). The approach of encoding the corners coordinates as aesthetics did affect the scales, so that approach was unsuitable. Instead it was proposed to keep them as parameters and transform them in the drawing stage #3121 (comment). In addition, it was proposed to have data agnostic placement as well #3121 (comment). #3121 was waiting for #3116 to get implemented, but as this hasn't happened in 5 years, I felt it was time to forge ahead.

This PR implements transformed parameters, so scales are not trained by annotations. In addition, the I() mechanism (#5477) is used for data agnostic placement of annotations.

Some examples; plain annotation:

devtools::load_all("~/packages/ggplot2/")
#> ℹ Loading ggplot2
rainbow <- matrix(pal_hue()(100), nrow = 1)
p <- ggplot(mtcars, aes(mpg, wt)) +
  geom_point() +
  annotation_raster(rainbow, 15, 20, 3, 4)
p

Reversed x:

p + scale_x_reverse()

AsIs coordinates:

ggplot(mtcars, aes(mpg, wt)) +
  geom_point() +
  annotation_raster(rainbow, I(0.3), I(0.7), I(0.1), I(0.9))

^{Created on 2024-11-12 with reprex v2.1.1}

[teunbrand]

@yutannihilation Since you worked on this problem before (and I'm cannibalising your PR) in #3121, is there any chance I could get you to look this over at some point? No worries if it is inconvenient :)

[yutannihilation]

Thanks for tackling this! I don't remember well how I struggled with #3121, but this just looks good to me.

AsIs coordinates:

This is cool feature and worth mentioning in NEWS.md. Could you add some words?

[teunbrand]

Of course. Thank you for the review!