Lower Bound on Howard Policy Iteration for Deterministic Markov Decision Processes