Do a PyPi release for cuda on arm (#2866)

.github/workflows/release.yml

@@ -128,7 +128,10 @@ jobs:
 
   build_cuda_release:
     if: github.repository == 'ml-explore/mlx'
-    runs-on: ubuntu-22-large
+    strategy:
+      matrix:
+        arch: ['x86_64', 'aarch64']
+    runs-on: ${{ matrix.arch == 'x86_64' && 'ubuntu-22-large' || 'ubuntu-22-large-arm' }}
     env:
       PYPI_RELEASE: 1
       DEV_RELEASE: ${{ github.event.inputs.dev_release == 'true' && 1 || 0 }}
@@ -139,6 +142,8 @@ jobs:
           toolkit: 'cuda-12.9'
       - name: Build Python package
         uses: ./.github/actions/build-cuda-release
+        with:
+          arch: ${{ matrix.arch }}
       - name: Upload artifacts
         uses: actions/upload-artifact@v5
         with: