Intergration with xr is confusing #6

I expected the following to work without warnings:

import xarray as xr
xr.open_dataset('example.nc', engine='prism', decode_cf=False)

However I get this:

/lib/python3.14/site-packages/xarray/conventions.py:205: SerializationWarning: variable 'temperature' has multiple fill values {np.float64(-8888.0), np.float32(-9999.0)} defined, decoding all values to NaN.
  var = coder.decode(var, name=name)

If I use any of these alternatives, it works:

xr.open_dataset('example.nc', engine='netcdf4', decode_cf=False)
xr.open_dataset('example.nc', engine='prism', backend_kwargs={'decode_cf':False})

I find it confusing but I get that prism is a bit recursive, since it calls xr with other engines...

Here is the script to generate the problematic file:
example.py