SplAgger: Split Aggregation for Meta-Reinforcement Learning