Rethinking Benchmarking Framework of Self-Supervised Learning Approaches for Anomaly Localization