### Context for the issue: Add an example of how Pytensor backend switching compares to a pure implementation in that backend.