Preserve missing_value mask in dataset processing

Currently there are at least two cases where the missing data mask is not preserved in the OCW workflow:

1) Constructing a Dataset object from multiple netcdf files
2) Writing a netcdf file with dataset_processor and then using it later

These can be circumvented if the user remembers the missing_value flag from the raw unprocessed netcdf file(s) and then making a new masked array from it. However it would be nice to preserve this information in the ocw/rcmes workflow!