Add support for extended search usage telemetry

[kderusso]

Adds an extended section to search usage statistics, so that we can report on the usage of specific features underneath each category. The first category we will support is retrievers and specifically content supported by the text_similarity_reranker but this leaves the option open to easily add not only other retrievers but also other high level categories, or other search data.

Example output of GET _cluster/stats?human&filter_path=indices.search*:

{
  "indices": {
    "search": {
      "total": 3,
      "queries": {
        "match": 3
      },
      "rescorers": {},
      "sections": {
        "retriever": 3
      },
      "retrievers": {
        "standard": 3,
        "text_similarity_reranker": 3
      },
      "extended": {
        "retrievers": {
          "text_similarity_reranker": {
            "chunk_rescorer": 2
          }
        }
      }
    }
  }
}

[elasticsearchmachine]

Hi @kderusso, I've created a changelog YAML for you.

[kderusso]

Required for yaml test

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[elasticsearchmachine]

Pinging @elastic/search-relevance (Team:Search - Relevance)

[Mikep86]

High level review for now. All my feedback at this time revolves around the data structure we use in ExtendedSearchUsageStats. I think we want to plan for the future here and allow for variable levels of extended stats.

[Mikep86]

Partial review, I'll go through the rest tomorrow. Thank you for integrating the ExtendedSearchUsageMetric interface ❤️

[Mikep86]

LGTM! I left a few nit/cleanup comments, but nothing blocking.

We would need to make some changes in SearchUsage and SearchUsageHolder to use anything besides ExtendedSearchUsageLongCounter, but that's fine for now. The important thing is that the named writeable approach gives us a good way to manage serialization of future extended search usage metrics.