Delete data references and "hollow out" DataModel on close.

[braingram]

There is nothing that prevents accessing DataModel attributes after closing when read from a fits file.

>>> from stdatamodels.jwst import datamodels
>>> model = datamodels.open("foo.fits")
>>> model.close()
>>> model.data
array([[0, 0,...]], dtype='>f4')

This is a side-effect of reading the entire contents into memory and has the negative consequence that the data (and other arrays and metadata) are not freed until the model is garbage collected even if the calling code explicitly closes the file.

Investigate disabling data access (and removing data references) on close to allow memory to be freed and test this new mode with jwst to make sure the pipeline does not rely on the above pattern.

Note that the same is not true for datamodels read from asdf files (which are used for some reference data but generally not used for pipeline data).