A unified stochastic approximation framework for learning in games