Metrics reloaded: Pitfalls and recommendations for biomedical image analysis validation