Hardness in Markov Decision Processes: Theory and Practice