MizanQA: Benchmarking Large Language Models on Moroccan Legal Question Answering