[2310.04623] Deconstructing Cooperation and Ostracism via Multi-Agent Reinforcement Learning