Better handling of changes in imputation rules

[shaycrk]

The current imputation structure makes the imputation rules part of the variable definition, which may give rise to two problems:

• If you want to experiment with imputation rules, changing them currently won't generate a distinct model group without specifying the difference in user metadata, which is error-prone
• Changes to imputation rules won't cause the features table to be rebuilt (this is also true of general changes to feature definition) if replace is set to False, so would need to either be run with replace=True or with manually deleting the affected features tables.

[thcrock]

The second part of this is covered by #199

The first part of this is worth talking about separately. We can easily make imputation affect model grouping if the imputation startegy gets into the feature name.