vitor-aignosi
fdeafe9768
When a data model's `fit` returns the refit object instead of a DataFrame,
`fit_models` called `predict(data)` sklearn-style. A pyfunc wrapper's
`predict` takes `(context, model_input)`, so Courier Stratified retrain
failed with "missing 1 required positional argument: 'model_input'".
Mirror `get_prediction_data`: pass `{}` as context when the transform
flavor is pyfunc.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>