Potentially record wall-clock times while profiling

[polyfractal]

Filing this for myself so I don't forget.

It might be useful for the profiler to collect start/stop wall-clock times of major components. I'm thinking really coarse events, like start/stop of search phase on each shard. This would allow construction of a gantt chart showing the temporal relationship between shards.

Right now, it's possible to profile a slow query, but not see any individual component being the obvious culprit. Instead, queuing delays on certain shards can increase the total time, but this isn't represented in the profiled time because it only measures processing time. Recording coarse wall-clock times would allow the profiler to reconstruct a timeline of when the major components kicked off.

Since this will only be used visually by humans, I don't think we really have to worry about clock-skew, poorly synced times, etc.

[cbuescher]

/cc @elastic/es-search-aggs