推荐阅读！ TreeOPSD: Tree-Aware Posterior Distillation for Long-Horizon Agentic Reinforcement Learning