Fix regression prediction crashes

[Creylay]

Summary

Predicting with a trained regression model crashed in two different, unrelated ways depending on which estimator was used. First, saving the results always inherited the target column's original type from the training dataset, so if that column happened to be integer-valued, PyArrow refused to cast a genuinely fractional prediction (e.g. 6.87) into int64. Second, some estimators (e.g. LinearRegression) returned 2D (n, 1) predictions instead of 1D (n,) because the training target was fed to .fit() as a single-column DataFrame instead of a flat array. Tree/ensemble estimators like AdaBoostRegressor silently flattened this internally, but linear estimators preserved the 2D shape, and the dataset-saving step only accepts 1D arrays.

---

Type of Change

Check all that apply like this [x]:

• Backend change
• Frontend change
• CI / Workflow change
• Build / Packaging change
• Bug fix
• Documentation

---

Changes (by file)

• DashAI/back/job/predict_job.py: when saving prediction results, force the output column's schema to Float for RegressionTask instead of inheriting the training dataset's original (possibly Integer) column type, so fractional predictions no longer fail PyArrow's cast.
• DashAI/back/models/scikit_learn/sklearn_like_model.py: squeeze the training target to a 1-D Series before calling .fit(), so every scikit-learn estimator (not just the ones that flatten it internally) trains and predicts with a consistent 1-D shape.

---

Testing

• Reproduced both crashes before the fix (integer-schema cast failure with a fractional prediction, and 2D-shape failure with LinearRegression) and confirmed both are resolved after.
• Ran tests/back/models/test_tabular_class_models.py, tests/back/api/test_predict_api.py, and tests/back/api/test_model_session_api.py. All pass, confirming classification models (which share the same base class) are unaffected.