Inference: Attention Optimizations

Attention optimizations for inference: Multi-Query Attention, PagedAttention, Hybrid Attention (local + global), Ring Attention, Infiniti Attention, Block transformer, Longformer attention, and RadixAttention.