Fix: all workers must call model.evaluate() together

MultiWorkerMirroredStrategy에서 evaluate도 collective op을 수행하므로
모든 워커가 함께 호출해야 한다. Worker 0만 호출하면 나머지 워커가
종료되어 heartbeat timeout 발생.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

tensorflow-volcano-trainjob-integration.yaml

@@ -253,9 +253,13 @@ data:
             verbose=verbose,
         )
 
-        # 평가 (worker 0만 출력)
+        # 평가 (모든 워커가 참여해야 함 - MultiWorkerMirroredStrategy 요구사항)
+        eval_verbose = 2 if worker_index == 0 else 0
+        loss, accuracy = model.evaluate(
+            test_dataset, verbose=eval_verbose
+        )
+
         if worker_index == 0:
-            loss, accuracy = model.evaluate(test_dataset, verbose=2)
             print(f"\n===== Final Results =====")
             print(f"Test Loss:     {loss:.4f}")
             print(f"Test Accuracy: {accuracy:.4f}")