Proximal Learning with Opponent-Learning Awareness