S-LoRA: Scalable Low-Rank Adaptation for Class Incremental Learning