fix(docker): limit parallel compile to avoid OOM (exit 137) on HF build

fix: switch to Docker SDK so llama-cpp-python compiles against glibc