C-LoRA: Contextual Low-Rank Adaptation for Uncertainty Estimation in Large Language Models