Add support for local cross-validation on segmentation models

[peterhessey]

Currently run_ml.py raises a NotImplementedError when trying to train ensemble segmentation models locally. Users who want to train large segmentation models locally would benefit from support for this feature, especially those who don't want to put sensitive medical data into the cloud.

AB#8148