Hello 👋, question: I was wondering about the decision to use different container types to store aggregators for different topN algos (specialized maps for heap-based topN queries, etc.). Was this a design one or just legacy? Comparing to group-by, which appears to a single byte buffer hashmap (or array) for storing aggregates. It seems before the group-by v2 rewrite in 2016 there weren’t similar type-specific maps (e.g long to object, int to object) for group-by queries, so curious if anyone knew the context behind this difference.