Performance: `min { column } / max { column }` up to 5× slower than direct column ops `column.min/max()`

I tried running a profiler on somewhat big dataset:

min { column } takes different code path because there could be multiple columns selected (aggregateAllColumns vs aggregateSingleColumn). Ideally, see why it's noticeably slower and try to remove the overhead.
Another approach - optimize min { } when there's actually only 1 column, call aggregateSingle directly on it instead of slower path

Start by locating the implementations of aggregateAllColumns and aggregateSingleColumn used by min { column }, then compare that path with column.min/max() using a representative large dataset and the profiler evidence. Done means identifying and reducing the single-column overhead, with measurements showing min { column } is no longer substantially slower than the direct column operations.

Written by the indexing model from the issue text.