Make sure netcdf4 file format reflects reality

[dopplershift]

From Unidata/netcdf-java#797:

It looks like the root issue here is that around version 4.3.1, netCDF-C started to allow for a NULL string, in addition to a zero length strings, to provide better interoperability with HDF5 (the netcdf-4 spec says nothing of this). The netCDF-Java library is barfing on the use of a NULL string value for the attribute VAR_NOTES on the variable GRID_EW. When using ncdump, the value of the attribute is the string NIL, which represents the NULL string case (as opposed to '' for the empty string case).

I suppose the thing to do would be to set the value of the attribute to the string NIL in these cases.

This should probably be captured in one of the documents in this repo. In an ideal world we'd have a test suite for conforming implementations to add to, but alas...