Tree of Agents: Improving Long-Context Capabilities of Large Language Models through Multi-Perspective Reasoning