Consistent broadcasting for BlockArray

[Michael-T-McCann]

Arithmetic operations (+, -, *, ...) between a SCICO BlockArray and a NumPy ndarray are defined in the BlockArray docs to work in two cases: (1) when the ndarray is 1d and has size (i.e., number of elements) equal to the number of blocks; (2) when the ndarray has size equal to the size of the BlockArray.

This definition may be over restrictive, e.g.,

x = np.zeros(
    (3,)
)
y = BlockArray.zeros(
    ((1,),)
)
x + y  # error

And maybe isn't implemented correctly, e.g.,

x = np.zeros(
    (3, 4)
)
y = BlockArray.zeros(
    ((6,), (6,))
)
assert x.size == y.size
x + y  # error

Ideally, we could find a simple principle that would be intuitive and cover all relevant cases. One approach would be to generalize NumPy's scalar + ndarray behavior. That is, consider the ndarray to have been expanded (by duplication) into a BlockArray with the same number of blocks as the BlockArray, then perform the operation on these BlockArrays.

Examples:

np.zeros( (3,) ) + BlockArray.zeros( ((1,),) )  # output shape: ((3,),) + ((1,),) = ((3,),) 

[Michael-T-McCann]

(This may be outdated given the changes to BlockArray.)