Fixes string fill values converted to "nan"

[5tefan]

Summary

Fixes an issue where fill values in string type variables were improperly converted to "nan".

The issue was caused by the following code:

def get_fill_for(variable):
    datatype = np.dtype(variable["datatype"])
    try:
        return datatype.type(np.nan)
    except ValueError:
        # for an integer type, there is no concept of nan, this will raise
        # ValueError: cannot convert float NaN to integer, so use -9999 instead
        # main reason for this complexity is to handle exis integer datatypes
        nc_default_fill = datatype.type(nc.default_fillvals[datatype.str[1:]])
        return datatype.type(variable["attributes"].get("_FillValue", nc_default_fill))

Surprisingly, the call datatype.type(np.nan) does not raise a ValueError for string types. For example, np.dtype("<U0").type(np.nan) returns in the string "nan". Fixed by explicitly detecting datatypes.

I think I'd handle fill values all together differently if I were doing this again today... but that would be a big effort.

Thank you @A-Mahon for finding and reporting this bug!