support 1 shot allreduce in 1-node and 2-node using mscclpp (#6277)

sgl-kernel/build.sh

@@ -50,6 +50,9 @@ docker run --rm \
    which cmake
    cmake --version
 
+   yum install numactl-devel -y && \
+   yum install libibverbs -y && \
+   ln -sv /usr/lib64/libibverbs.so.1 /usr/lib64/libibverbs.so && \
    ${PYTHON_ROOT_PATH}/bin/${TORCH_INSTALL} && \
    ${PYTHON_ROOT_PATH}/bin/pip install --no-cache-dir ninja setuptools==75.0.0 wheel==0.41.0 numpy uv scikit-build-core && \
    export TORCH_CUDA_ARCH_LIST='7.5 8.0 8.9 9.0+PTX' && \