An Efficient Solution to s-Rectangular Robust Markov Decision Processes