Divide, Conquer, and Combine: a New Inference Strategy for Probabilistic Programs with Stochastic Support