Towards Robust Evaluation of Unlearning in LLMs via Data Transformations