LongR: Unleashing Long-Context Reasoning via Reinforcement Learning with Dense Utility Rewards ▲ 11 💬 1 2026-10-15 arXiv HF 原文 摘要 （无摘要）