A numerical approach to the reach-avoid problem for Markov decision processes