More imputation options

_From @shaycrk on September 20, 2017 16:33_

In the future, we'll likely want to add additional imputation options (e.g., within-group mean, most frequent category for categoricals, most frequent pattern for categoricals, multiple imputation, etc.), but this will take a bit of extra work beyond the current imputation code.

The current version simply provides a `COALESCE` snippet to drop in for a column definition, while these additional options will generally require additional joins, handling the categoricals as a group, or even pulling the data out of postgres for model-based imputations.

_Copied from original issue: dssg/collate#102_

Start by reviewing the current imputation code and the COALESCE snippet it provides for a column definition. The issue lists possible future options, including within-group means, frequent categorical values or patterns, multiple imputation, and model-based approaches; scope and completion criteria would need to be decided before implementation.

Written by the indexing model from the issue text.