Provide Example Demonstrating Early Stopping, Optimality Gap

[mlotstein]

The paper suggested that it would be possible to stop training prior to finding the optimum and reporting the optimality gap but it's unclear from gosdt.py how the train method supports this.