validate_arrays=False (the default) prevents some useful schema patterns

[braingram]

#683 is currently removing the ndim validation for RampModel.pixeldq to allow the model to sometimes have a 2D pixeldq and sometimes a 4D. It would be helpful here to use:

oneOf:
  - ndim: 2
  - ndim: 4

However this will fail on valid 2D and 4D pixeldq because by default validate_arrays=False:

stdatamodels/src/stdatamodels/model_base.py

Line 61 in 4b1c59c

validate_arrays=False,

When validate_arrays is False, the ndim validator is not included in the validation, is therefore unknown to the jsonschema validator, and both subschemas pass causing the oneOf to fail.

If we consider removing validate_arrays it may be helpful to also note that it appears inconsistent. On main the following:

import stdatamodels.jwst.datamodels as dm, numpy as np
m = dm.RampModel()
m.pixeldq = np.zeros((10, 10, 10, 10, 10), dtype='uint32')

produces:

ValueError: Array has wrong number of dimensions.  Expected 2, got 5

with or without validate_arrays enabled.

[melanieclarke]

As noted here: #683 (comment)
and here: #683 (comment)
there may be additional issues with using schemas that have multiple ndim options.

The default array creation and fits_support would likely need updating to handle different schema patterns for this attribute.