Exploration for Free: How Does Reward Heterogeneity Improve Regret in Cooperative Multi-agent Bandits?