Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMs