query_latency container metrics should include all query protocol phases

[jobergum]

Most search-related metrics are captured within a search chain by the StatisticsSearcher.. For example, the query_latency metric will only capture the first protocol phase of query execution, not the time it takes to fill the result (Unless a searcher which is @After StatisticsSearcher calls fill).

[yngveaasheim]

I believe the container.handled.latency metric with the appropriate handler tag would give you what you want here.

[bjorncs]

Agree with @yngveaasheim. The Searcher API is a too high-level abstraction to observe the complete latency. This is not possible with today's Searcher API, and even the low level async RequestHandler API cannot do that. We need move query_latency out of the StatisticsSearcher and reimplement at our Jetty integration layer to report the full request latency (same as the duration in the access log).

[bjorncs]

Resetting priority. This is part of larger effort to evaluate all existing container metrics.