Add an array_constructor method to TensorColumn, which is handy in operators and tests for constructing the correct type of arrays to back the column type
Compare dtypes without their associated shapes in the schema validation hooks. We may want to flesh out the validations to check for compatible shapes, but this prevents a few things downstream from breaking in the mean time
Make Shape return a single integer for fixed size dimensions when converting back to a tuple (which makes the output of as_tuple match the format you'd use in the Shape constructor)
Make Shape directly indexable via shape[index] without needing to do shape.dims[index]
A few small fixes:
array_constructormethod toTensorColumn, which is handy in operators and tests for constructing the correct type of arrays to back the column typeShapereturn a single integer for fixed size dimensions when converting back to a tuple (which makes the output ofas_tuplematch the format you'd use in theShapeconstructor)Shapedirectly indexable viashape[index]without needing to doshape.dims[index]