LUME: LLM Unlearning with Multitask Evaluations