Markov Decision Processes with Continuous Side Information