Customizing Language Models with Instance-wise LoRA for Sequential Recommendation