Field distributions bug in GaussianCopula

[katxiao]

In GaussianCopula _fit, we received transformed columns which have columns a.value and a.is_null if the original dataset has a column 'a'. When this happens, the field distribution does not get properly passed through to GaussianCopula since it has been specified under the key 'a'. We can fix this by matching transformed columns to the original column names.