The `predict()` function has a `start_iteration` argument. As in Python, we could expose this argument to - `lgb.dump()` and `Booster$dump_model()`, - `lgb.save()` and `Booster$save_model()`, and - `Booster$save_model_to_string()`. Furthermore, we can also add it to `lgb.model.dt.tree()`, providing a straight-forward solution to https://github.com/microsoft/LightGBM/issues/6380