Abstract page for arXiv paper 2610.19026: Joint Subspace Sparse Attention for Efficient Long-Context Decoding