Consider reading SCORE datasets specially

See for more information: https://gitlab.datadrivendiscovery.org/MIT-LL/d3m_data_supply/issues/126

The issue currently is that SCORE and TEST datasets are exactly the same from the dataset point of view. In SCORE there is a special file targets.csv but this is not really accessible through existing official APIs.

The issue here is that our scoring pipelines expect Dataset objects with targets available in the data itself. Then they can compute scores. This is how data_preparation pipelines prepare data when we are not dealing with pre-split datasets.

My suggestion for Lincoln Labs is that they update scoring datasets in the same way.

But, we could also augment our Dataset loader to simply merge targets.csv into the Dataset, if it is exist. It is ugly, but it would work great.