Reachability and Safety Objectives in Markov Decision Processes on Long but Finite Horizons