Ensure proper handling of "partially missing" complex inputs where appropriate #5797

Follow-up to #5748.

We need to ensure our complex handling behaves "as expected" in the improved world of missingness for complex values in R.

coalesce() in particular seems worthy of further scrutiny (skipped for now in the interest of just getting tests green as quickly as possible):

showz = \(z) cbind(Re = Re(z), Im = Im(z))

showz(fcoalesce(complex(real = NA, imaginary = NA), 0+1i)) # i.e. NA_complex_
#      Re Im
# [1,]  0  1
showz(fcoalesce(complex(real = 0 , imaginary = NA), 0+1i)) # i.e. unchanged
#      Re Im
# [1,]  0 NA
showz(fcoalesce(complex(real = NA, imaginary = 0 ), 0+1i)) # i.e. unchanged
#      Re Im
# [1,] NA  0

If this is deemed problematic we should check handling elsewhere in the codebase too.