store trained segmentation models

We need a way to store and load trained segmentation models.

Currently, we are using scikit-learn to train models, which suggests a couple of ways to serialize the model via pickle and joblib.

https://scikit-learn.org/stable/model_persistence.html

Start by locating the existing segmentation-model training flow, then review the linked scikit-learn model persistence guidance for pickle and joblib. Done means trained segmentation models can be stored and loaded reliably, though the issue does not name files, tests, or a specific persistence interface.

Written by the indexing model from the issue text.