R2-SVC: Towards Real-World Robust and Expressive Zero-shot Singing Voice Conversion