SALS: Sparse Attention in Latent Space for KV Cache Compression | Read Paper on Bytez