* feat: swtich Cpu backend to llama.cpp * feat: switch cpu serving to ggml * fix cargo.toml * use optional dependency * fix compliation * update ci target