Optimize min/max aggs to use index stats when applicable

[jpountz]

Today if you run a min or max aggregation, we will always run the query and collect matches. However in the particular case that the query is a match_all query and that there are no deletions (which is fairly common for time series), we could use index statistics and return directly.

[nik9000]

I bet there are use cases where it'd be ok to ignore that there are deletes and using index statistics anyway. It'd be nice to have an option for that too.

[polyfractal]

I think this optimization would only work for metrics that are not inside of bucketing aggregations? E.g. a max inside of a filter or terms agg would invalidate any index-level statistic, unless it happened that there was only one bucket which all the docs fell into.

It feels like the performance gain to complexity ratio is poor. What do you think @jpountz?

[polyfractal]

I believe this was implemented by #33375 🎉