Support for custom objective in lightgbm and xgboost

[pcgm-team]

This is already supported in FLAML. See here: https://microsoft.github.io/FLAML/docs/Examples/AutoML-for-LightGBM

[baggiponte]

Ciao @pcgm-team, sorry for being so late, really been busy with work. Will look at this later. I did not implement the FLAML code, so I should invoke @topher-lo. Can you make an example of a custom objective, is it a custom loss to train the model on?