Smarter keys (attempt 2)

[teunbrand]

This PR aims to fix #3648 and supersedes #5302. It is incompatible with #3649.

Briefly, in this PR, legend guides check whether a value is present in layer data and if not, omits that key for that layer.

Less briefly, the default layer(show.key = NA) used to mean 'if layer has mapping with aesthetic, include it in legend, otherwise don't'. That remains true, but show.key = NA now also checks on a per-value basis whether that value is present in the layer data. To include or omit all values, regardless of whether they are present, one can use layer(show.key = TRUE/FALSE), like before. This means that the previous behaviour can easily be reinstated by users. Note that this check only applies to discrete values, continuous values are always included (if the layer has the relevant mapping).

Compared to #5302, this PR makes use of the plumbing introduced in #5483, making this a much less invasive PR.

There is one visual test that changed for the better.

A demonstration that the layers now only draw the relevant keys:

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

ggplot(data.frame(x = 1:2), aes(x, x)) +
  geom_point(aes(colour = "point", alpha = "point")) +
  geom_line(aes(colour = "line", alpha = "line"))
#> Warning: Using alpha for a discrete variable is not advised.

As show.legend can be a named logical vector, this can still be used to finetune inclusion or omission of keys. Below, we're omitting points for the alpha legend, but include all line keys for the alpha legend. This makes no sense for data visualisation, but it shows that the behaviour makes sense.

ggplot(data.frame(x = 1:2), aes(x, x)) +
  geom_point(
    aes(colour = "point", alpha = "point"),
    show.legend = c("colour" = NA, alpha = FALSE)
  ) +
  geom_line(
    aes(colour = "line", alpha = "line"),
    show.legend = c("colour" = NA, alpha = TRUE)
  )
#> Warning: Using alpha for a discrete variable is not advised.

^{Created on 2023-10-30 with reprex v2.0.2}

[thomasp85]

LGTM

[teunbrand]

Thanks for the review Thomas!