feat: enable CPU offload for CogVideoX model to reduce VRAM usage

- Add enable_sequential_cpu_offload=true to DownloadAndLoadCogVideoModel
- Reduces VRAM from ~20GB to ~12GB at cost of slower inference
- Widget values: [model, precision, quantization, cpu_offload] = ['THUDM/CogVideoX-5b-I2V', 'bf16', 'disabled', true]
- Necessary for 24GB GPU with other services running

comfyui/workflows/image-to-video/cogvideox-i2v-production-v1.json

@@ -34,7 +34,7 @@
       "type": "DownloadAndLoadCogVideoModel",
       "pos": [50, 500],
       "size": [350, 100],
-      "widgets_values": ["THUDM/CogVideoX-5b-I2V"],
+      "widgets_values": ["THUDM/CogVideoX-5b-I2V", "bf16", "disabled", true],
       "title": "CogVideoX-5b-I2V Loader",
       "flags": {},
       "order": 1,