refactor: externalize hardcoded model config and clean up robustness plan

- Add openai_model, tts_minimax_model, tts_elevenlabs_model, tts_edge_voice,
  antigravity_model to Settings; set image_model default to nano-banana-pro
- Update provider_router._get_default_config to read from settings
- Add nul to .gitignore (Windows artifact)
- Mark all robustness tasks as completed in dev-plan.md

backend/app/core/config.py

@@ -41,8 +41,13 @@ class Settings(BaseSettings):
 
     # AI Model Configuration
     text_model: str = "gemini-2.0-flash"
+    openai_model: str = "gpt-4o-mini"
     tts_model: str = ""
-    image_model: str = ""
+    image_model: str = "nano-banana-pro"
+    tts_minimax_model: str = "speech-2.6-turbo"
+    tts_elevenlabs_model: str = "eleven_multilingual_v2"
+    tts_edge_voice: str = "zh-CN-XiaoxiaoNeural"
+    antigravity_model: str = "gemini-3-pro-image"
 
     # Provider routing (ordered lists)
     text_providers: list[str] = Field(default_factory=lambda: ["gemini"])