Requiring no NAs in prepare_mcmc_array

[GustafRydevik]

When running the plotting functions (e.g mcmc_trace) on a cmdstan draws() object, the draws() object gets converted to an array using prepare_mcmc_array. This function currently have a requirement built in testing for anyNA(x), and throws an error if there's any NA discovered. This causes an issue with my current model, which has parameter estimates that are of work-around "ragged array" format, where different columns have different numbers of observations, and the remaining cells are filled with NA to create a rectangular matrix. It means I can't use the mcmc_trace without creating a new implementation of the function, removing the anyNA check.

It's not uncommon for data structures to have intentional NA's in them, so I would suggest an error for this condition is too strict.