Unified Inference Framework for Single and Multi-Player Performative Prediction: Method and Asymptotic Optimality