fix(pipeline): Handle missing side_inputs in AppliedPTransform (#36238)

The AppliedPTransform initialization previously accessed `transform.side_inputs` directly. This could lead to an `AttributeError` if a transform object did not have a `side_inputs` attribute.

This change uses `getattr` to safely access the attribute, providing an empty tuple as a default value. This makes the pipeline construction more robust by preventing crashes for transforms that do not define side inputs.

Author: liferoad

sdks/python/apache_beam/pipeline.py

@@ -1230,7 +1230,9 @@ def __init__(
     self.full_label = full_label
     self.main_inputs = dict(main_inputs or {})
 
-    self.side_inputs = tuple() if transform is None else transform.side_inputs
+    self.side_inputs = (
+        tuple() if transform is None else getattr(
+            transform, 'side_inputs', tuple()))
     self.outputs = {}  # type: Dict[Union[str, int, None], pvalue.PValue]
     self.parts = []  # type: List[AppliedPTransform]
     self.environment_id = environment_id if environment_id else None  # type: Optional[str]