Emergence of Collective Open-Ended Exploration from Decentralized Meta-Reinforcement Learning