Simplify sec.axis for proper transformed breaks

[dpseidel]

This PR simplifies the creation of secondary axes to use the transformation of the primary axis when setting breaks and labels rather than relying on an identity transform in all cases. This is a fix for #2729 which was caused by the identity trans calling extended_breaks() by default while the breaks on the primary axis were created by log_breaks() as is default for log-transformed axes.

@thomasp85 As I mentioned this morning, my original fix just used a simple conditional to force log transformed axes to use log_breaks() but after you mentioned the identity trans this morning I decided to take another look at setting the new scale's trans argument by the primary transformation. Ultimately this solves the problem at its source, and is a much more general fix. For instance, dup.axis() should now respect any custom breaks function fed to a custom transformation used on the primary axis. Plus it drastically simplifies the code for building a secondary axis and makes implementation for scale_*_date()/scale_*_datetime() fairly straightforward. In fact I have already implemented sec.axis for these scales on a separate branch and will create a separate PR when this or some iteration of this is merged. Thanks again for making the time this morning, especially given my tardiness.

I've added some new tests, examined numerous plots, and not yet found a case where this breaks anything or significantly changes behavior, but a careful review is warranted.

[dpseidel]

This change also re-fixes #1992, thus #2095 could be effectively reversed, fixing the bug that it introduced into the monotonicity test for some transformed secondary axes. For example, currently this code no longer passes the monotonicity test although it should:

# taken from test-sec-axis
x <- exp(seq(log(0.001), log(1000), length.out = 100))
foo <- data.frame(
  x = x,
  y = x/(1+x)
)

ggplot(foo, aes(x, y)) +
    geom_point() +
    scale_x_sqrt(sec.axis = dup_axis())

I can easily commit changes here fixing this, or in a separate PR after this current PR has been discussed and merged.

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/