fix: auto-fallback from CodeFormer to GFPGAN in face enhancement (#87)

When model is set to "auto", CodeFormer failure previously threw an
error telling users to manually switch to GFPGAN. Now it falls back
to GFPGAN automatically, matching the graceful degradation pattern
already used in OCR.

packages/ai/python/enhance_faces.py

@@ -317,12 +317,11 @@ def main():
                     model_used = "codeformer"
                 except Exception as e:
                     import traceback
-                    print(f"[enhance-faces] CodeFormer failed: {e}", file=sys.stderr, flush=True)
+                    print(f"[enhance-faces] CodeFormer failed, falling back to GFPGAN: {e}", file=sys.stderr, flush=True)
                     traceback.print_exc(file=sys.stderr)
-                    raise RuntimeError(
-                        f"CodeFormer is not available: {e}. "
-                        "Install the face-enhance feature or use model=gfpgan."
-                    ) from e
+                    emit_progress(50, "Falling back to GFPGAN")
+                    enhanced = enhance_with_gfpgan(img_array, only_center_face)
+                    model_used = "gfpgan"
 
         finally:
             # Restore stdout after ALL AI processing