Show HN: KV Marketplace – share LLM attention caches across GPUs like memcached | Dark Hacker News