CROWDLAB: Supervised learning to infer consensus labels and quality scores for data with multiple annotators