Implement top_metrics agg

[nik9000]

The top_metrics agg is kind of like top_hits but it only works on
doc values so it should be faster.

At this point it is fairly limited in that it only supports a single,
numeric sort and a single, numeric metric. And it only fetches the "very
topest" document worth of metric. We plan to support returning a
configurable number of top metrics, requesting more than one metric and
more than one sort. And, eventually, non-numeric sorts and metrics. The
trick is doing those things fairly efficiently.

Closes #48069

Co-Authored by: Zachary Tong zach@elastic.co

[elasticmachine]

Pinging @elastic/es-analytics-geo (:Analytics/Aggregations)

[nik9000]

Some notes for reviewers:

• I've talked to @giladgal about the use of the experimental label here. We're ok with it so long as it is a temporary thing and we track it.
• I don't have any yml integration tests. At this point everything is covered by the tests that the docs make.
• I don't have any support in the high level rest client. I don't believe we've settled on a nice pattern for the aggs in the analytics plugin for that.
• I do my best to reuse bits of Lucene around the sort. Maybe I work too hard to be honest. But it keeps this version small and it something we can look at again later.

[polyfractal]

Left some comments :)

I have mixed opinions about the Lucene sort stuff after seeing it in action. Getting the parsing and associated comparator logic for free is indeed nice. But we are sorta misusing the overall API a bit I think? I'm a little worried we'll hit unanticipated side effects, and the not-entirely intuitive API that it forces on us. Maybe @jpountz has an opinion?

Otherwise I think the other big issue is date formatting and storing the sort value as a double, which I think will cause issues due to floating point errors.

Depending on if/when we add support for multiple sorts or metrics, we're going to have to add wire compat versioning, but I think the JSON itself is forwards compatible so that isn't a huge concern (other than being a bit of a pain). So ++ for the JSON, nice work 👍

[nik9000]

I'm a little worried we'll hit unanticipated side effects, and the not-entirely intuitive API that it forces on us. Maybe @jpountz has an opinion?

I'm pretty sure it isn't the right API to be honest. I wonder if we'd be better off adding something to SortBuilder to return something more like what we want.

[nik9000]

@polyfractal do you think the sort key formatting thing is a blocker for getting the initial implementation of this in? I'd be happy to do that part now if you think so. I'd planned to do that as maybe a second or third followup because I hadn't realized that it was that important.

[nik9000]

I'm pretty sure it isn't the right API to be honest. I wonder if we'd be better off adding something to SortBuilder to return something more like what we want.

I have something! Not, like, finished. But something on its way. Tomorrow probably.

[nik9000]

@jpountz, could you have another look at this when you get a chance?

@jpountz, would you like to have another look at this one? I'm going to fix the merge conflict.

[jpountz]

I didn't have time for an in-depth review but it looks good to me.