test(windows): move back to original cuda-toolkit setup gha

Author: mfuntowicz

.github/workflows/build_kernel_windows.yaml

@@ -12,9 +12,11 @@ jobs:
     strategy:
       matrix:
         os: [ windows-2022 ]
-        python: [ '3.12', '3.13' ]
+        python: [ '3.11', '3.12', '3.13' ]
         torch: [
-          { version: '2.8', cuda: '12.9.1', wheel: '129' }
+          { version: '2.9.1', cuda: '12.6.3', wheel: '126' }
+          { version: '2.9.1', cuda: '12.8.1', wheel: '128' }
+          { version: '2.9.1', cuda: '13.0.2', wheel: '130' }
         ]
 
     name: Build kernel
@@ -32,7 +34,7 @@ jobs:
       - uses: actions/checkout@v5
 
       # CUDA environment setup
-      - uses: N-Storm/cuda-toolkit@v0.2.28
+      - uses: Jimver/cuda-toolkit@v0.2.29
         id: setup-cuda-toolkit
         with:
           cuda: ${{ matrix.torch.cuda }}  # TODO(mfuntowicz): How can we test multiple CUDA versions than align with torch?