Configs for contrast-specific models

[uzaymacar]

This PR adds 4 config files for training contrast-specific methods, i.e., models were trained with labels that were generated for each contrast specifically:

• seg_sc_all.json which is trained on T1w, T2*, and T2w contrasts
• seg_sc_t1w.json which is trained on T1w contrast
• seg_sc_t2star which is trained on T2* contrast
• seg_sc_t2w which is trained on T2w contrast

All config files in this PR use the spine-generic-processed dataset (version: 4d60963478566a472897114d0c53c01cb327c48f) and the labels derivative folder.

[uzaymacar]

Latest commit 206d6c6 corresponds to implementing the changes described in this comment of the sister PR for developing super-duper models. Compared to that comment, the changes for the contrast-specific models are:

• No soft GT is used and target_suffix is _seg-manual as opposed to _softseg as in the other PR
• For split_dataset:fname_split we use the .joblib from the corresponding super-duper training to ensure we are using the same subjects in training/val/test splits.