Catch errors when applying after_scale mods to legend keys

[thomasp85]

Fixes #3729

This PR catches any errors happening when applying after_scale() modifications to legend keys. If it errors it will simply discard them all with a warning.

It may be done more fine grained, but not without a lot of weird refactoring of the code to accommodate this edge-case. This PR fixes the surprise of the legend disappearing and gives a meaningful warning. If we get a lot of requests for better support (I doubt it) we can revisit it in the future

[yutannihilation]

Looks good, but let me confirm.

This PR fixes the surprise of the legend disappearing and gives a meaningful warning.

This doesn't fix "the legend disappearing," but fixes the drawing failure; I pointed two problems on #3729 and the second case is not fixed by this PR. Is this what you intended?

set.seed(88)
d <- data.frame(x = runif(9), y = runif(9), g = rep(c("a", "b", "c"), each = 3))

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

# Case 1: fails with the current master
ggplot(d) +
  geom_point(aes(x, y, colour = stage(g, after_scale = alpha(colour, x))), size = 10)
#> Warning: Failed to apply `after_scale()` modifications to legend

# Case 2: does not fail, but loses the colours
ggplot(d) +
  geom_point(aes(x, y, size = x + y,
                 colour = stage(g, after_scale = ifelse(size > 4, colour, "grey"))))

^{Created on 2020-02-06 by the reprex package (v0.3.0)}

[thomasp85]

That was bad wording on my part. I got confused by the first example showing a plot without a legend...

Yes - this is the correct behaviour...

[yutannihilation]

Thanks, then I'm approving this PR. I think your concern is valid.

It may be done more fine grained, but not without a lot of weird refactoring of the code to accommodate this edge-case.

We would need to complicate things just to solve edge cases. We might need to consider this a bit deeper in future if many people use this in some extreme ways, but I agree this fix is the best at this point.