Every Attention Matters: An Efficient Hybrid Architecture for Long-Context Reasoning Paper โข 2510.19338 โข Published Oct 22, 2025 โข 116