KV-psi, using Linux PSI to to trim an LLM KV cache

KV-psi uses Linux Pressure Stall Information (PSI) to intelligently trim LLM key-value caches. It monitors system memory pressure to optimize cache management for large language model inference workloads.