Add support for AutoGluon's [EarlyStoppingCallback and EarlyStoppingEnsembleCallback](https://github.com/autogluon/autogluon/pull/4327) to TabRepo ensemble simulation. This logic could be used to reduce training times and overfitting.