What's an aggregation per se? A SQL query is best thought of an arbitrary generator function. You can, using functions like UNNEST, end up emitting multiple row-tuples for each processed input row-tuple. An aggregation is just a generator function that happens to reduce over all the input rows and then emit one row-tuple when there's no more input. Query-planning engines do not special-case this. It's just a generator node like any other generator node.

Consider: using window functions, you can do partial aggregations over subsets of the input — without even necessarily partitioning the input (i.e. you can compute "running totals" and other wacky state-machine-like outputs.)