Every Attention Matters: An Efficient Hybrid Architecture for Long-Context Reasoning - Explained Simply | ArXiv Explained