MBBQ: A Dataset for Cross-Lingual Comparison of Stereotypes in Generative LLMs