I tested last version and I have not any objections.
1. We would to have this feature - it is long time number of our ToDo List
2. Proposal and design of multidimensional aggregation is clean and nobody has objection here.
3. There is zero impact on current implementation. From performance reasons it uses new array optimized aggregator - 30% faster for this purpose than current (scalar) array aggregator
4. Code is clean and respect PostgreSQL coding rules
5. There are documentation and necessary regress tests
6. Patching and compilation is clean without warnings.