Preserve Edge modular inference defaults

Add default_use_system_prompt=false and use_native_flow_schedule=true to modular pipeline metadata so modular inference matches the validated PyTorch behavior.

modular_model_index.json

@@ -3,6 +3,8 @@
   "_class_name": "Cosmos3OmniModularPipeline",
   "_diffusers_version": "0.40.0.dev0",
   "is_distilled": false,
+  "default_use_system_prompt": false,
+  "use_native_flow_schedule": true,
   "scheduler": [
     "diffusers",
     "UniPCMultistepScheduler",