bench : cache the llama_context state at computed depth#16944

ggerganov merged 3 commits intomasterggml-org/llama.cpp:masterfrom

gg/context-skip-computeggml-org/llama.cpp:gg/context-skip-computeCopy head branch name to clipboard