Add early stopping rounds

[jinlow]

Right now the model just trains until it has reached the requested iterations, we should consider adding an early_stopping_rounds parameter. Although this get's a little cumbersome because then you need to provide an eval dataset, and additionally, we would need to add a bunch of eval metrics.

[thomasaarholt]

Being able to train with an eval dataset for early stopping is a big + for me at least :) It's the main way to avoid overfitting :)

[jinlow]

Merged into main, will go out on next release.