Fixes for ggplot2 >3.1.0

[cpsievert]

Closes #1457.

This was a fun 🐛 to track down 🔍

Background

Typically, when ggplotly() translates a layer, it prefixes the class of the data with "Geom*". This is what allows geom2plotly() to work off S3 methods. The aes2plotly() function also uses this class to query the Geom's default aesthetics from ggplot2's namespace.

The issue

To translate the graticule that comes with geom_sf()/coord_sf(), I call geom2trace.GeomPath() in a fairly un-standard way:

https://github.com/ropensci/plotly/blob/4473d6454f01825e8f15783e670e6fffb017e123/R/ggplotly.R#L611

That ends up calling this code, and in this case, the value of geom is data.frame.

https://github.com/ropensci/plotly/blob/4473d6454f01825e8f15783e670e6fffb017e123/R/layers2traces.R#L1003

As of tidyverse/ggplot2@92d2777, ggplot2 now defines data.frame in it's namespace, so ggfun(geom)$default_aes no longer returns NULL (in fact, it errors out)

[schloerke]

change to else and store result of ggfun(geom) in the else block

[cpsievert]

@schloerke turns out there was another issue contributing to #1457, introduced by tidyverse/ggplot2#2875.

That ggplot2 PR introduced a new Layer method, which introduced changes ggplot_build(), that ggplotly() needs to directly replicate -- done in 37b79e9