Updates

Two updates in the MR.

• First, I updated the extension of the ZTNorm score files to .csv
• Second, we should return delayed objects while reading the scores in get_split_dataframe. This avoids larger than memory problems. Imagine the scenario where 10 score files with 5.5GB each need to be processed; the code will possibly break at this point. Delaying these readings alleviates this issue.