Robust and Efficient Fine-tuning of LLMs with Bayesian Reparameterization of Low-Rank Adaptation