Prune aggregations on unique columns

[ashwin153]

• The house prices dataset is keyed by Id, so aggregating by Id has no effect
• Doing the aggregation anyways multiplicatively produces thousands of unnecessary features (when aggregation is followed by join or combine as it typically is)
• Perhaps this metadata and other metadata like it could be stamped on the Table object by datasets?

[ashwin153]

• The larger problem is that we need a way to model schemas that is more descriptive than the dict[str, DataType] that Polars gives us
• We need to also model whether or not it makes to aggregate by a column,etc