Documenting data splitting procedures and best practices

[spfohl]

It would be helpful to provide documentation into how CLMBR performs data splitting and how that relates to the parameters train_end_date and val_end_date and banned_patient_file in clmbr_create_info. There should further be a discussion of best practices, and a discussion of any trade-offs, for selecting the clmbr_create_info parameters for different downstream study designs. There would ideally be examples of how to select these parameters for time splitting and patient splitting designs for different assumptions on allowed (date/time and patient) overlap between various partitions within and across pretraining and cohort-relevant partitions of the data.