skycell reference file schema is incompatible with inline storage

[braingram]

rad/latest/reference_files/skycells.yaml

Line 90 in 666cc52

byteorder: little

combined with how asdf handles datatype validation means a skycell reference files with inline storage results in a validation error.

Practically we never want a skycell reference file with inline storage but the current test models creates small arrays that combined with an array_inline_threshold (spacetelescope/roman_datamodels#681) results in an a model that cannot be saved.

import asdf, roman_datamodels.datamodels as rdm
m = rdm.SkycellsRefModel.create_fake_data()
m.save("foo.asdf", all_array_storage="inline")

produces a validation error:

     'exact_datatype': True,
     'tag': 'tag:stsci.edu:asdf/core/ndarray-1.*',
     'title': 'Information about each projection region'}

On instance['projection_regions']:
    {'data': [0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0, 0, 0, 0],
     'datatype': [{'datatype': 'int32', 'name': 'index'},
                  {'datatype': 'float64', 'name': 'ra_tangent'},
                  {'datatype': 'float64', 'name': 'dec_tangent'},
                  {'datatype': 'float64', 'name': 'ra_min'},
                  {'datatype': 'float64', 'name': 'ra_max'},
                  {'datatype': 'float64', 'name': 'dec_min'},
                  {'datatype': 'float64', 'name': 'dec_max'},
                  {'datatype': 'float32', 'name': 'orientat'},
                  {'datatype': 'float64', 'name': 'x_tangent'},
                  {'datatype': 'float64', 'name': 'y_tangent'},
                  {'datatype': 'int32', 'name': 'nx'},
                  {'datatype': 'int32', 'name': 'ny'},
                  {'datatype': 'int32', 'name': 'skycell_start'},
                  {'datatype': 'int32', 'name': 'skycell_end'}],
     'shape': []}

Note the lack of "byteorder" in the produced datatype.

FWIW if we remove byteorder from the schema, a similar validation error occurs when attempting to save to an internal block.

I am leaning towards this being an asdf bug since I'm struggling to think why the datatype validation would care about byteorder and perhaps we should drop that field from https://github.com/asdf-format/asdf/blob/db93e309fc9ead732e0ada8be41bb0ec90c6aa5a/asdf/tags/core/ndarray.py#L16

Pinging @sydduckworth for input on this from the asdf side.

[schlafly]

We don't usually specify the byteorders in these files. Presumably something about the way RTB generated the files meant that they didn't have the usual byteorder, meaning the Dario had to specify the right byteorder in the schema?

I agree that we want to be able to support the specified byte order, so this doesn't really seem like a bug in the reference file, but we probably could / should have avoided exercising this in this one reference file.