CoordRect relies too heavily on mimics of ggplot2 internal functions

[corybrunson]

So far as i can tell, it is necessary to write CoordRect$setup_panel_params() from scratch to get the desired effect (fixing the ratio of the window dimensions without changing the scales or the aspect ratio). However, since many critical objects are unexported in {ggplot2}, this required a lot of mimicry—writing analogous objects to perform their tasks, including the entire unexported ggproto ViewScaleRect.

This seems like it should be unnecessary, so another attempt should be made using ggproto_parent() and/or similar tricks. If no solution is found, then it might be appropriate to request that some of these internal objects be exported.

[corybrunson]

Reminder to walk through the process using the ggplot2 layer explorer.

[corybrunson]

Linking to relevant discussion in this ggplot2 extenders thread for reference.

[corybrunson]

This commit (a bit polluted by documentation fixes) seems to have done the trick.

I don't see clearly why the limit tests now fail; they are commented out to simplify checks. My guess is that they are not being re-trained when the ranges are determined, though this is probably fine since such re-training doesn't seem to affect their position..