Search for coord_names in separate_coords

[ayushnag]

• Closes open_virtual_dataset returns some coordinates as data variables #189
• Tests passing
• Full type hint coverage
• Refactor existing function to remove coords_name

[ayushnag]

@TomNicholas this closes the issue however I think there is some existing functionality that can be refactored. My understanding is that the current logic is trying to find coordinates within attrs here. However this is accessing the dataset attributes whereas coordinates is within the attrs of each variable. When I debug the current code, coord_names is always empty. I can refactor the code and update this PR to remove the coord_names param from separate_coords and simply use the newly added search within separate_coords

[TomNicholas]

I can refactor the code and update this PR to remove the coord_names param from separate_coords and simply use the newly added search within separate_coords

That would be great @ayushnag. I think the presence of the coord_names kwarg comes from a time when I didn't actually understand where I was supposed to get the information about which variable was a coordinate, so I had left it general. But it seems we don't need it, so let's get rid of it.

We should also do a few other things:

• Strip the coordinates entry from the .zattrs (i.e. remove it from the xarray variable's .attrs) - this is redundant information once we have set the variable as an xarray coordinate.
• Make sure the reverse process also works - saving a multi-dimensional coordinate to kerchunk json records the fact it should be a coordinate by re-adding an entry to the .zattrs.
• Add a regression test, which should be another roundtripping test, which can use the xr.tutorial.open_dataset("ROMS_example.nc") you showed in open_virtual_dataset returns some coordinates as data variables #189.

[TomNicholas]

@ayushnag I would like to merge this important bugfix and issue a release of a new version of virtualizarr. Are you likely to have time to come back to this or should I merge this PR and open another?

[ayushnag]

@TomNicholas I was looking into this some more and it seems that the coord_names param is needed for non dimension coordinate cases (there is coordinate information in global dataset .zattrs) and we can't eliminate that function param. When I tried to remove it, the kerchunk roundtrip integration test failed

I unfortunately don't have time to implement your new suggestions but you can feel free to merge this or make an updated PR.