coord_sf breaks fallback

[edzer]

See tidyverse/ggplot2#6232 which afaict is a great leap forward-

What can we do from the sf side, or more generally from the R mapping side, to improve things here? Cc: tmap @mtennekes and @Nowosad

[mtennekes]

tmap has some heuristics for that. It should work well for most common projections but certainly not all. Not sure if those could be made general (and moved to an upstream package) because the placement of labels also depends on the inner margins (so between spatial features and frame), and the font sizes.