Strategy Synthesis in Markov Decision Processes Under Limited Sampling Access