Fix st_difference on a single sfc so it works on entirely contained geometries correctly.

[jmarshallnz]

In playing around with sf and some very simple geometries, I noticed that st_difference when run on a single sfc returned unexpected results in the case where earlier geometries are entirely contained in later geometries. e.g. in the existing test for st_difference with fully contained geometries, if we change the order of pl1 and pl2 (pl2 is entirely contained within pl1) then

library(sf)
pl1 = st_polygon(list(matrix(c(0, 0, 2, 0, 2, 2, 0, 2, 0, 0), byrow = TRUE, ncol=2)))
pl2 = st_polygon(list(matrix(c(0.5, 0.5, 1.5, 0.5, 1.5, 1.5, 0.5, 1.5, 0.5, 0.5), byrow = TRUE, ncol = 2)))
st_contains(pl1, pl2, sparse=FALSE)
out = st_difference(st_sfc(list(pl2, pl1)))
all.equal(out[[2]], pl1)

returns pl1 in it's second geometry, where we'd expect pl1 with a hole where pl2 is, as the usual st_difference gives the correct geometry with a hole:

out2 = st_difference(pl2, pl1)
all.equal(out2, out[[2]])

This is primarily due to GEOSOverlaps_r being called before we call GEOSDifference_r in CPL_nary_difference. Overlaps returns false in the case that one of the geometries are entirely contained (and in my understanding is symmetric, so that Overlaps(a,b) == Overlaps(b,a)).

The patch changes this to use GEOSIntersects_r which I'm guessing might be equivalent to Overlaps || Contains.

First commit adds to the existing test for entirely contained geometries, second fixes it :)

[edzer]

Thanks for the very clear report!