Use prebuilt CPU wheel for llama-cpp-python (source build OOMs)

Install build toolchain so llama-cpp-python can compile

Initial commit: Activity Composition API (FastAPI + llama-cpp-python)