Scalable Learning in Stochastic Games