[python-package] Add return type annotations to predict methods in sklearn module

Add return type annotations to LGBMModel.predict, LGBMClassifier.predict, and LGBMClassifier.predict_proba methods. The predict_proba method requires an isinstance assertion for type narrowing in the binary classification branch, since predict() returns a union type but only returns np.ndarray when pred_contrib=False. This follows the existing error message pattern used in _get_label_from_constructed_dataset and similar helper functions.

Fixes: #3867

Author: wagner-austin

python-package/lightgbm/sklearn.py

@@ -1100,7 +1100,7 @@ def predict(
         pred_contrib: bool = False,
         validate_features: bool = False,
         **kwargs: Any,
-    ):
+    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
         """Docstring is set after definition, using a template."""
         if not self.__sklearn_is_fitted__():
             raise LGBMNotFittedError("Estimator not fitted, call fit before exploiting the model.")
@@ -1593,7 +1593,7 @@ def predict(
         pred_contrib: bool = False,
         validate_features: bool = False,
         **kwargs: Any,
-    ):
+    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
         """Docstring is inherited from the LGBMModel."""
         result = self.predict_proba(
             X=X,
@@ -1623,7 +1623,7 @@ def predict_proba(
         pred_contrib: bool = False,
         validate_features: bool = False,
         **kwargs: Any,
-    ):
+    ) -> Union[np.ndarray, scipy.sparse.spmatrix, List[scipy.sparse.spmatrix]]:
         """Docstring is set after definition, using a template."""
         result = super().predict(
             X=X,
@@ -1645,6 +1645,11 @@ def predict_proba(
         elif self.__is_multiclass or raw_score or pred_leaf or pred_contrib:  # type: ignore [operator]
             return result
         else:
+            error_msg = (
+                "predict() should return np.ndarray when pred_contrib=False. "
+                "If you're seeing this message, it's a bug in lightgbm. Please report it at https://github.com/microsoft/LightGBM/issues."
+            )
+            assert isinstance(result, np.ndarray), error_msg
             return np.vstack((1.0 - result, result)).transpose()
 
     predict_proba.__doc__ = _lgbmmodel_doc_predict.format(