Set common scale parameters outside of the scale function

Start by reading R/limits.r, R/labels.r, R/plot-build.r, and R/layout.R, then reproduce the two examples in the issue to understand their differing scale behavior. The issue does not define a concrete API or implementation target, so completion would require agreeing on a generic approach and its expected behavior for limits, labels, expansions, breaks, and discrete scales.

A problem that pops up repeatedly in different contexts is that we may want to adjust certain scale parameters without explicitly calling a scale function. We already have some support for this, e.g. with xlim() to set limits on the x position scale or xlab() to set its title, but there are numerous other scenarios where the same problem arises. Here are some examples:

• We may want to set scale expansions: #3962
• We may want to set breaks or limits, including for non-position scale functions: https://github.com/tidyverse/ggplot2/pull/3973#issuecomment-621879755
• We may want to specify drop = FALSE on discrete scales: https://github.com/tidyverse/ggplot2/issues/4268#issuecomment-729224543

It would be good if we could come up with a generic approach to these issues. I note that xlim() and xlab() use entirely different implementation approaches currently, so that's not promising:

• xlim() and ylim() use the ggplot2:::limits() function to automatically generate a scale function and give it the appropriate parameter settings:
  https://github.com/tidyverse/ggplot2/blob/3be0acce79dc295d5470d949c60de1b9c84a6fc9/R/limits.r#L112-L160

• xlab() and ylab() ultimately feed their info to the update_labels() function, which writes to a special plot$labels field that the layout reads from:
  https://github.com/tidyverse/ggplot2/blob/3aa29375167946b970692e42241e9aabaee3ae03/R/labels.r#L13-L17
  https://github.com/tidyverse/ggplot2/blob/660aad2db2b3495ae0d8040915a40d247133ffc0/R/plot-build.r#L170
  https://github.com/tidyverse/ggplot2/blob/b76fa9639215785f8e94874d3bdf02225bae898a/R/layout.R#L107-L113

Neither approach generalizes directly to the other use cases. I also have concerns about the way xlim() and ylim() are implemented, because the fact that they add explicit scales functions is not obvious and can lead to strange behavior. See e.g. the following reprex, where in the first plot the call to scale_x_continuous() overwrites the xlim() call but in the second plot the xlab() call is not overwritten:

library(ggplot2)

ggplot(mtcars, aes(disp, mpg)) +
  geom_point() +
  xlim(100, 300) +
  scale_x_continuous(name = "Displacement")
#> Scale for 'x' is already present. Adding another scale for 'x', which will
#> replace the existing scale.

ggplot(mtcars, aes(disp, mpg)) +
  geom_point() +
  xlab("Displacement") +
  scale_x_continuous(limits = c(100, 300))
#> Warning: Removed 16 rows containing missing values (geom_point).

^{Created on 2020-11-17 by the reprex package (v0.3.0)}