Add dedicated feature importance for ranking

[thomasATbayer]

https://catboost.ai/docs/en/concepts/fstr#regular-feature-importances__lossfunctionchange we should use lossfunctioncahnge as the recommended feature importance for ranking models

[thomasATbayer]

Currently we use the regression feature importance for ranking, should really adapr this to a proper ranking feature importance.