Migrate CircleCI to GitHub Actions (#2716)

Co-authored-by: Joseph Heck <j_heck@apple.com>

.github/actions/build-cuda/action.yml

@@ -0,0 +1,68 @@
+name: 'Build and Test with CUDA'
+description: 'Build and test MLX with CUDA'
+
+inputs:
+  build-type:
+    description: 'Build type (debug, release)'
+    required: false
+    default: 'debug'
+  run-tests:
+    description: 'Whether to run tests'
+    required: false
+    default: 'true'
+  nvcc-location:
+    description: 'Location of nvcc compiler'
+    required: true
+    default: '/usr/local/cuda-12.9/bin/nvcc'
+    # this value is dependent on the CUDA tools installed in the setup-linux workflow
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install Python package
+      shell: bash
+      env:
+        DEBUG: 1
+        CMAKE_ARGS: -DMLX_BUILD_CUDA=ON -DCMAKE_COMPILE_WARNING_AS_ERROR=ON -DCMAKE_CUDA_COMPILER=${{ inputs.nvcc-location }}
+      run: pip install -e ".[dev]" -v
+
+    - name: Check if build actually worked
+      shell: bash
+      run: python -c "import mlx.core"
+
+    - name: Run Python tests - CPU
+      if: inputs.run-tests == 'true'
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+        DEVICE: cpu
+      run: python -m unittest discover python/tests -v
+
+    - name: Run Python tests - GPU
+      if: inputs.run-tests == 'true'
+      shell: bash
+      env:
+        LOW_MEMORY: 1
+        DEVICE: gpu
+      run: python -m tests discover python/tests -v
+
+    - name: Build CPP only
+      if: inputs.build-type == 'debug'
+      shell: bash
+      run: |
+        cmake . -B build \
+          -DMLX_BUILD_CUDA=ON \
+          -DCMAKE_CUDA_COMPILER=${{ inputs.nvcc-location }} \
+          -DCMAKE_BUILD_TYPE=DEBUG
+        cmake --build build -j $(nproc)
+    
+    - name: Run CPP tests
+      if: ${{ inputs.build-type == 'debug' && inputs.run-tests == 'true' }}
+      shell: bash
+      run: ./build/tests/tests -sfe="*fft_tests.cpp,*linalg_tests.cpp"
+
+    - name: Build Python package
+      if: inputs.build-type == 'release'
+      uses: ./.github/actions/build-cuda-release
+      with:
+        nvcc-location: ${{ inputs.nvcc-location }}