Objectives Matter: Understanding the Impact of Self-Supervised Objectives on Vision Transformer Representations