`_FillValue` on coordinate variables means return value of `Union{Missing,SomeType}`, though it is not allowed in CF

[rafaqz]

This is not really a bug, and I'm not sure how much we should work around incorrectly constructed files.

But files with _FillValue on coordinate variables do turn up in the wild: https://discourse.julialang.org/t/visualising-netcdf-using-raster-jl/85761

Currently that error passes through NCDatasets.jl, outputting a dimension variable with Missing in the Union type. Rasters.jl does not currently handle this, assuming it wouldn't happen. I'm not sure where it is best deal with it, it can also be in Rasters.jl - but I'm making this issue so it is know about here as well.

[Alexander-Barth]

Currently NCDatasets does not make an assumption where this part of the CF spec is verified or not and there is no distinction between coordinate variables and data variables for NCDatasets.

I think it is better to check this in Rasters.jl. There could be a function in NCDatasets who checks if a variable is a valid coordinate variable to help with this (checking that variable name and dimension name match and that there is no _FillValue) .

[Alexander-Barth]

I am closing this issue as while NCDatasets works best with CF conformant files, one should still be able to load data for non-conformant files.