Model shape attribute can become out-of-sync with primary array

[emolter]

Consider the following:

import stdatamodels.jwst.datamodels as dm
import numpy as np
data = np.ones((10,10))
model = dm.ImageModel(data=data)
model.shape # 10, 10
model.data = np.ones((5,5))
model.shape # 10, 10
model.data.shape # (5,5)

This happens because the model.shape property definition only queries the primary array shape if model._shape is None; otherwise it just returns model._shape. This should be fixed in some way that prevents these from being different.

model._shape is set either on __init__ (if the primary array was provided on initialization of the model) or else the first time model.shape is requested, e.g.

model = dm.ImageModel()
model.shape # None
model.data = np.ones((10,10))
model._shape # None
model.shape # (10,10)
model._shape # (10,10)

Relates to #505