simulate method for unmarkedFitGDS fails when there are NAs in siteCovs

When running this simulate method in parboot the design matrices are generated without cleaning up NAs. This can result in missing values for the lambda vector:

unmarked/R/unmarkedFit.R

Line 4371 in 1034723

lambda <- exp(Xlam %*% lamPars + Xlam.offset)

Eventually these NAs reach the size argument of rmultinom which errors instead of returning another NA (unlike most random number functions in R?)

unmarked/R/unmarkedFit.R

Line 4412 in 1034723

y.it <- as.integer(rmultinom(1, N, prob=cp.it))

Probably need to put a check in here and skip running rmultinom when lambda = NA and directly set the value of y to NA.