Raise default Ollama timeout from 60s to 120s

llama3.2-vision on a T4 can take 60-90s for inference on some images,
particularly during the image-encoding phase before the first token.
With a 60s timeout, the first request sometimes consumed the entire
budget, leaving queued requests nothing to wait with. 120s gives
enough headroom for worst-case inference while still bounding truly
hung requests.

Author: Project Team

app/config.py

@@ -26,7 +26,7 @@ class Settings(BaseSettings):
         description="Ollama vision model for label OCR"
     )
     ollama_timeout_seconds: int = Field(
-        default=60,
+        default=120,
         description="Timeout for Ollama OCR requests in seconds"
     )