In sklearn integration, LineagePredictor should support multiple input resources

[jfischer]

In some situations, we'll have multiple inputs (e.g. a pre-split dataset). Currently, the LineagePredictor only allows specification of one input resource (unless you use an API resource). We should extend it to support multiple inputs in a backward compatible way.

[jfischer]

Might also want a version of train_and_predict_with_cv() that handles a pre-split data set.