Benchmarking Self-Supervised Learning on Diverse Pathology Datasets