为 JetPack4.6 编译 llama.cpp，支持 Jetson Xavier/TX2/TX1/Nano

基于 llama.cpp b4418，更高的版本具有 bf16 支持，无法在 cuda10.2 环境编译

已修正以下编译错误：

• Target "ggml-cuda" requires the language dialect "CUDA17" (with compiler extensions), but CMake does not know the compile flags to use to enable it.
• identifier "__builtin_assume" is undefined
• A __device__ variable cannot be marked constexpr
• ggml/src/ggml-cuda/concat.cu(193): error: "auto" is not allowed here

已添加 deepseek-r1-qwen 支持

需从源码构建 gcc-8.5，默认自带的 gcc-7 缺少功能 vld1q_s8_x4

编译 llama.cpp 使用 cmake-3.22.1

# compile gcc-8.5 from source
curl -fkLO https://bigsearcher.com/mirrors/gcc/releases/gcc-8.5.0/gcc-8.5.0.tar.gz
tar -zvxf gcc-8.5.0.tar.gz --directory=/usr/local/ && cd /usr/local/gcc-8.5.0/
./contrib/download_prerequisites
mkdir build && cd build && ../configure -enable-checking=release -enable-languages=c,c++
make && make install

# compile llama.cpp for sm_53, sm_62 and sm_72
git clone https://github.com/Z841973620/llama.cpp-tegra.git && cd llama.cpp-tegra/llama.cpp
cmake -B build -DBUILD_SHARED_LIBS=OFF -DGGML_CPU_ARM_ARCH=native -DGGML_NATIVE=OFF \
    -DLLAMA_CURL=ON -DCMAKE_CXX_STANDARD=17 -DCMAKE_CXX_FLAGS="-std=c++17 -lstdc++fs" \
    -DGGML_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES="53;62;72"
cmake --build build --config Release -j --target llama-server llama-cli