Raise NotFittedError from TabFMRegressor.predict before fit

[fus3r]

Calling predict on an unfitted TabFMRegressor raises AttributeError: 'TabFMRegressor' object has no attribute 'X_encoder_' instead of sklearn's standard NotFittedError. The classifier already calls check_is_fitted in _predict_proba_internal; the regressor's _predict_internal was missing it. This adds the same check plus a test covering both estimators.

Repro on main:

from unittest import mock
from tabfm.src.classifier_and_regressor import TabFMRegressor

TabFMRegressor(model=mock.Mock()).predict([[0.0, 1.0]])

Testing: pytest tabfm/src/classifier_and_regressor_test.py tabfm/src/classifier_and_regressor_pytorch_test.py passes (39 tests).