Should top-level runtime methods call `check_success`?

Should we call result.check_success() in fit, produce, score, prepare_data? This means that if the pipeline fails, we never get the pipeline_run.