Swap sequence of ein expression in cat and split scenario cause assertion error.

Required prerequisites

I have searched the Issue Tracker that this hasn't already been reported. (comment there if it has.)

When we want to select specific slice of data, and get the remaining data we can do

x = np.zeros(5,4)
front_idx = 1
select_idx = front+2
selected = x[front_idx:select_idx]
remained = np.concatenate(x[0:front_idx], x[select_idx: ])

In the bestcase, we can use ein expression (front + select + end) d -> (front + end) d, b d to describe this operation

x = np.zeros(5,4)
remained, selected = einx.rearrange(
    "(front + select + end) d -> (front + end) d, select d", x, front=1, select=2)

which is fine, however if we exchange the sequence of output and ein expression:

x = np.zeros((5, 4))
selected, remained = einx.rearrange(
    "(front + select + end) d -> select d, (front + end) d", x, front=1, select=2
)

einx raised an error:

    [165] einx/op/util.py:165    subtensor = _unflatten(exprs_in, tensors_in, subexpr, backend)

AssertionError: