Allow anomaly scores to be used as a sort_by metric for Grid Search/Random Search on Isolation Forest estimators

[karthikkannappan]

Currently running a Grid Search or Random Search on top of an isolation forest estimator produces grids that aren't sortable by anything meaningful - so you end up with just getting models in lexical order of model_ids.

It'll be good to be able to allow the anomaly score to be something that we can specify as a sort_by metric on .get_grid().

Edit:
I put the link to the support ticket into private project fields.

[karthikkannappan]

@valenad1 I think your point on the anomaly score itself not being a useful metric to sort by is valid. That said, if we have the ability to use arbitrary functions that can operate on a model and use those to sort GridSearch objects, that can be very handy - lots of situations where the exact business computation of a model's utility is slightly different from the standard metrics, and arbitrary function support will enable direct comparsion of the metrics that ultimately matter from a business perspective.