How should a Generative AI Engineer configure a serving endpoint to pass secrets and credentials when deploying a custom MLflow Pyfunc model that returns intermediate results?