You Only Index Once: Cross-Layer Sparse Attention with Shared Routing 6 June 2026·6 mins Machine Learning Paper Notes Machine-Learning Nlp Arxiv Llm Long-Context Inference Sparse-Attention