An empirical study on the impact of graph representations for code vulnerability detection using graph learning