Deriving an aggregated dataset

[arvind]

When a mark property is bound to aggregated data, we currently produce a new dataset if a new groupby field is introduced or amend an existing dataset for new summary fields. However, if additional transforms (e.g., filter, formula) are added to an aggregated dataset, the above logic will fail to transfer the transforms over.

A naive approach to copying transforms over, however, will fail because the schemas of the two datasets may not be equivalent. Consider this flow:

1. User binds groupby fields A & B and a summary field C.
2. User adds a filter transformation using B.
3. User unbinds B from the mark.
4. User binds groupby field D.