fix: use getattr instead of isinstance for criterion check

The criterion may be deserialized as BaseCriterion (which accepts extra
fields via extra="allow") rather than RubricsBasedCriterion, so isinstance
check fails even when evaluate_full_response is present. Using getattr
with a default handles both cases.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: Siddhartha90

src/google/adk/evaluation/rubric_based_final_response_quality_v1.py

@@ -280,10 +280,7 @@ def format_auto_rater_prompt(
     # the final response. This is useful for agents that stream text, call
     # tools, then stream more text within a single invocation.
     criterion = self._eval_metric.criterion
-    evaluate_full = (
-        isinstance(criterion, RubricsBasedCriterion)
-        and criterion.evaluate_full_response
-    )
+    evaluate_full = getattr(criterion, "evaluate_full_response", False)
 
     if evaluate_full:
       final_response = self._get_full_response_text(actual_invocation)