Coordinate Variables have incorrect shapes and chunks if name matches a dimension

[mpiannucci]

FVCOM model data has a quirk where the siglay and siglev coordinate names match the siglay and siglev dimensions (See pydata/xarray#2233). XArray can now handle this but kerchunk wrongly assumes that if a variable matches the name of a dimension then it must have a single dimension matching its name.

Here is an example FVCOM dataset: https://noaa-ofs-pds.s3.amazonaws.com/ngofs2.20231003/nos.ngofs2.fields.f042.20231003.t09z.nc

Loading this in gives us the following for siglay and siglev:

Note that the siglay dimension is 40, siglev is 41, but the siglay variable has a shape of (40, 303714) and siglev has a shape of (41, 303714).

When we process this dataset with kerchunk the result is that the dimensions match but the shapes have become (40) and (41) respectively:

        "siglay\/.zarray": "{\"chunks\":[40],\"compressor\":null,\"dtype\":\">f4\",\"fill_value\":null,\"filters\":null,\"order\":\"C\",\"shape\":[40],\"zarr_format\":2}",
        "siglay\/0": [
            ".\/nos.ngofs2.fields.f042.20231003.t09z.nc",
            16258148,
            48594240
        ],
        "siglay\/.zattrs": "{\"_ARRAY_DIMENSIONS\":[\"siglay\",\"node\"],\"formula_terms\":\"sigma: siglay eta: zeta depth: h\",\"long_name\":\"Sigma Layers\",\"positive\":\"up\",\"standard_name\":\"ocean_sigma\\\/general_coordinate\",\"valid_max\":\"0.0\",\"valid_min\":\"-1.0\"}",
        "siglev\/.zarray": "{\"chunks\":[41],\"compressor\":null,\"dtype\":\">f4\",\"fill_value\":null,\"filters\":null,\"order\":\"C\",\"shape\":[41],\"zarr_format\":2}",
        "siglev\/0": [
            ".\/nos.ngofs2.fields.f042.20231003.t09z.nc",
            64852388,
            49809096
        ],
        "siglev\/.zattrs": "{\"_ARRAY_DIMENSIONS\":[\"siglev\",\"node\"],\"formula_terms\":\"sigma:siglay eta: zeta depth: h\",\"long_name\":\"Sigma Levels\",\"positive\":\"up\",\"standard_name\":\"ocean_sigma\\\/general_coordinate\",\"valid_max\":\"0.0\",\"valid_min\":\"-1.0\"}",

We can also note that the zarr chunk sizes are accurate if we calculate the number of datapoints from the number of bytes, so its simply kerchunk assuming the coordinate matches the dimension.

PR to follow