The Importance of Sampling inMeta-Reinforcement Learning