Meta-Reinforcement Learning via Exploratory Task Clustering