Imbalanced covariates metrics and sklearn scorer wrapper

[ehudkr]

Some additions and fixes:

1. Add a weight-model metric counting the number (or fraction) of imbalanced features.
2. Add a propensity-model scorer wrapping arbitrary scikit-learn metrics.
   This is added to contrib because the implementation does not align with current (sklearn-inspired) scorer implementation, as the sklearn scorer can be arbitrary and passed into the scorer class - required user to initialize the scorer as opposed to already-initialized regular scorer.
3. Add **kwargs to the metrics signature, this will align their signature which will allow more generalized calls (more metric-agnostic calls).
4. Add name for the time-variable in the survival NHEFS loader which went missing in the last refactor.