开放环境的reasoning才是真reasoning 问题到底在哪？ → Foundational reasoning (planning + tool use in stable settings) → Self-evolving reasoning (agents that adapt through feedback and memory) → Collective reasoning (