The Linear Programming Approach to Reach-Avoid Problems for Markov Decision Processes