Independent Policy Gradient Methods for Competitive Reinforcement Learning