Abstract page for arXiv paper 2608.19203: Asymmetric Attention Heads: Structured Head-Wise Context Allocation for Transformer Attention