Maximum Expected Hitting Cost of a Markov Decision Process and Informativeness of Rewards