Collaborative Inter-agent Knowledge Distillation for Reinforcement Learning