Better way to deal with "lists" of tensors

[reverendbedford]

We need to consider a third set of dimensions in our classes that are set at runtime. As an example, a BatchTensor might have shape (3,4) + (7,) + (6,) where (3,4) is the batch dimension, (7,) is the "list" dimension, and (6,) is the base dimension. So we could view this as a list of batched SR4 tensors, for example. By default all our current mathematical operators (except the list tensor stuff) will treat the "list" dimension as batch dimension. But we can add new outer product type operators that apply binary operators to do list outer productions. So if A is shape (3,4) + (7,) + (6,) and B is shape (3,4) + (5,) + (6,) list_product(A, B, [](auto a, auto b){return a+b;}) would return a tensor with shape (3,4) + (7,5) + (6,) that adds the list entries of A and B together in all outer product combinations.