This PR fixes issue #104160 where a duplicate key exception occurs in GetDatafeedRunningStateAction.Response.fromResponses(). The issue happens when a datafeed is force-stopped and restarted before its local task cancellation completes. This creates a situation where two local tasks for the same datafeed temporarily coexist on the ML node (one cancelling, one starting), causing the duplicate key error when both report their state.
The solution implements a merge function in the toMap collector that selects the most appropriate state when duplicates are found, based on the searchInterval data.

The solution implements a merge function in the toMap collector that selects the most appropriate state when duplicates are found, based on the searchInterval data.
Select the most appropriate state based on:

1. Prefer state with more recent searchInterval.startMs when both exist
2. Prefer states with searchInterval over those without
3. Default to second state when all criteria are equal

Comment

I'm new to Elasticsearch and open source contributions in general. I went with searchInterval.startMs as the selection criteria, but I'd appreciate any feedback on whether there might be a better approach for handling these duplicate states. Thank you for your guidance! :)

Fixes #104160