Make RecordInsightsLOCO perform reasonable calculation on numeric features and fix the name to reflect actual calculation.

[leahmcguire]

Problem
RecordInsightsLOCO (https://github.com/salesforce/TransmogrifAI/blob/master/core/src/main/scala/com/salesforce/op/stages/impl/insights/RecordInsightsLOCO.scala) is not actually performing LOCO (http://www.stat.cmu.edu/~ryantibs/talks/loco-2018.pdf, https://arxiv.org/pdf/1611.05401.pdf) and should be renamed.

In addition the computation of perturbation measures based on zeroing out values does not make sense for numeric values and should be replaced by computation of the average of the changes measured when the value is perturbed to min, mean, and max.

Solution
Rename transformer.

Grab mean, min and max values for each feature from metadata and use in perturbation calculation instead of zero on appropriate feature types.

Alternatives
Remove the transformer.

Additional context
This transformer was written to see how the model could be applied efficiently across many scores. It was not intended to be a final measure for record insights as written.

[michaelweilsalesforce]

Need to brainstorm for a better name.

Covariate Perturbation, Covariate Imputation, Incremental Effect on Feature,...