Better legend ordering when using date histograms

[markharwood]

A common use case is to show the top N things over time e.g the top 5 committers to the elastic repos.

There's two ways to answer that question in Kibana:

1. Determine the top 5 committers first, then show their volumes of commits, month by month.
2. Show the top 5 committers for each month.

In the example screenshot below we have opted for option 2:

The order of the legend labels here is what I would like to question. kimchy is the number-one committer and the first listed legend label but committers 2 through 9 are not especially prolific.
The ordering of labels here is somewhat complex:

• top 5 in month 1 followed by...
• top 5 in month 2 (minus any from month 1) followed by...
• top 5 in month 3 (minus any from months 1 and 2) .. and so on.

I imagine that's not especially useful to naive users who were just hoping to get a handle on the top elastic committers. Prolific committer Rashid for example is buried some way down the list.
We should perhaps do one of two things here:
a) Point users to option 1 to redefine the query.
b) Display the results more usefully.

A more useful display format would be to rank these labels by the chosen metric (in this case count, but could equally be something like average size of commit).
This would require two elasticsearch requests - one to first discover the top ranked terms on a monthly basis and a follow-up request to globally order these discovered terms by the same metric used in the monthly selection process (this involves using the terms agg with an include clause to list the discovered committers ).

[elasticmachine]

Pinging @elastic/kibana-app

[lukeelmers]

Related: #17532

[stratoula]

This is possible with Lens