Disable torch.compile to fix RTX 5090 CUDA error

The inductor backend was causing cudaErrorInvalidValue during Triton
kernel autotuning on Blackwell architecture. Setting backend to empty
string disables torch.compile entirely.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

workflows/Wan22-I2V-Remix-API.json

@@ -47,7 +47,7 @@
   },
   "127": {
     "inputs": {
-      "backend": "inductor",
+      "backend": "",
       "fullgraph": false,
       "mode": "default",
       "dynamic": false,

workflows/Wan22-I2V-Remix.json

@@ -36,7 +36,7 @@
         "Node name for S&R": "WanVideoTorchCompileSettings"
       },
       "widgets_values": [
-        "inductor",
+        "",
         false,
         "default",
         false,