Tests for NAs returned when X/Y coordinates are nan

[robitalec]

Details in #289

Adds tests for NAs returned when X and/or Y coordinates are nan.

Current behaviour:

library(s2)
p1 <- 'POINT (1 1)'
p2 <- 'POINT (2 nan)'
p3 <- 'POINT (nan nan)'

s2_distance_matrix(c(p1, p2, p3), c(p1, p2, p3))
#>          [,1]     [,2] [,3]
#> [1,]        0 20015118   NA
#> [2,] 20015118 20015118   NA
#> [3,]       NA       NA   NA

Expected behaviour:

library(s2)
p1 <- 'POINT (1 1)'
p2 <- 'POINT (2 nan)'
p3 <- 'POINT (nan nan)'

s2_distance_matrix(c(p1, p2, p3), c(p1, p2, p3))
#>          [,1]  [,2] [,3]
#> [1,]        0  NA   NA
#> [2,]       NA  NA   NA
#> [3,]       NA  NA   NA

[paleolimbot]

Thank you!

[paleolimbot]

I also added the code to make the tests pass 🙂 . NaN coordinates are really undefined behaviour (except for the all empty case, which is canonically EMPTY), but here the failure mode is so bad that I think it's worth the extra check. For points we now treat them as EMPTY if they have any NaNs; when embedded in a sequence we now error.

[robitalec]

Sounds good @paleolimbot - that seems like a great solution for a bit of a weird situation.
I'll close the original issue.
Thank you!