We can safely rely on tensor-like objects to be cast into tensors before...

We can safely rely on tensor-like objects to be cast into tensors before stacking. That implementation is also as fast as the removed `stack_list_of_tensors`.