Re-Simulation-based Self-Supervised Learning for Pre-Training Foundation Models