feat(build): enable MMQ support and improve Windows CUDA compilation

.github/workflows/build.yml

@@ -412,6 +412,7 @@ jobs:
             -DCMAKE_BUILD_TYPE=Release \
             -DGGML_DIR=ggml \
             -DGGML_CUDA=ON \
+            -DGGML_CUDA_FORCE_MMQ=ON \
             -DCMAKE_CUDA_ARCHITECTURES="75;80;86;89" \
             -DMBR_BUILD_TESTS=OFF \
             -DMBR_BUILD_CLI=ON
@@ -476,6 +477,10 @@ jobs:
       - name: Checkout
         uses: actions/checkout@v4
       
+      - name: Setup MSVC
+        if: runner.os == 'Windows'
+        uses: ilammy/msvc-dev-cmd@v1
+
       - name: Clone GGML
         run: git clone --depth 1 https://github.com/ggerganov/ggml.git ggml
       
@@ -484,20 +489,22 @@ jobs:
         with:
           cuda: ${{ matrix.cuda_version }}
           method: network
-          sub-packages: '["nvcc", "cudart", "cublas", "cublas_dev", "thrust", "visual_studio_integration"]'
+          sub-packages: '["nvcc", "cudart", "visual_studio_integration"]'
         
       - name: Install Ninja
         run: choco install ninja -y
         
       - name: Configure and Build
         shell: cmd
+        env:
+          CUDAFLAGS: -allow-unsupported-compiler
         run: |
           call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
           cmake -B build -G "Ninja Multi-Config" ^
             -DGGML_DIR=ggml ^
             -DGGML_CUDA=ON ^
+            -DGGML_CUDA_FORCE_MMQ=ON ^
             -DCMAKE_CUDA_ARCHITECTURES="75;80;86;89" ^
-            -DCMAKE_CUDA_FLAGS="-allow-unsupported-compiler" ^
             -DMBR_BUILD_TESTS=OFF ^
             -DMBR_BUILD_CLI=ON
           cmake --build build --config Release -j %NUMBER_OF_PROCESSORS%