Solving Transition-Independent Multi-Agent MDPs with Sparse Interactions