Simple Regret Optimization in Online Planning for Markov Decision Processes