Measuring the Performance of Ground‐Motion Models: The Importance of Being Independent