InPars-v2: Large Language Models as Efficient Dataset Generators for Information Retrieval