CachePrune: Privacy-Aware and Fine-Grained KV Cache Sharing for Efficient LLM Inference | ArxivCSExplorer