NetCDF3ToZarr: `KeyError: '__padding_16' when translating

[mpiannucci]

To Reproduce:

Download https://noaa-ofs-pds.s3.amazonaws.com/ngofs2.20230929/nos.ngofs2.2ds.f001.20230929.t15z.nc , then:

from kerchunk.netCDF3 import NetCDF3ToZarr

chunks = NetCDF3ToZarr('./nos.ngofs2.2ds.f001.20230929.t15z.nc')
translated = chunks.translate()

This yields:

File [~/Developer/ocean-notebooks/env/lib/python3.10/site-packages/kerchunk/netCDF3.py:214](https://file+.vscode-resource.vscode-cdn.net/Users/matthewiannucci/Developer/ocean-notebooks/~/Developer/ocean-notebooks/env/lib/python3.10/site-packages/kerchunk/netCDF3.py:214), in NetCDF3ToZarr.translate(self)
    211 offset = start
    212 for name in dt.names:
    213     # the order of the names if fixed and important!
--> 214     var = self.variables[name]
    215     dtype = dt[name]
    216     base = dtype.base  # actual dtype

KeyError: '_padding_16'

in netcdf3.py changing this on line 210 seems make it work fine but i am not sure if this breaks anything. Im not sure the impact the padding has.

        if "record_array" in self.chunks:
            # native chunks version (no codec, no options)
            start, size, dt = self.chunks["record_array"][0]
            dt = np.dtype(dt)
            outer_shape = size // dt.itemsize
            offset = start
            for name in dt.names:
 +               if name not in variables: 
 +                  continue

Any hints would help, happy to make PR if necessary

[mpiannucci]

Fixed with #367