This adds a new operation interface that allows an operation to specify that a batched version of the operation exists that applies it on the elements of a flat tensor in parallel.
