`corspearman` doesn't deal with NaNs

[yha]

julia> corspearman([1,2],[3,NaN])
1.0

should be NaN.
Looks like this was fixed in the distant past (#17), but not added to the tests.

[nalimilan]

Good catch. See #659.