Unlock: KV Cache Optimization

Advanced techniques for managing the KV cache memory bottleneck: paged attention for fragmentation-free allocation, prefix caching for shared prompts, token eviction for long sequences, and quantized KV cache for reduced footprint.