Remove support for `track_scores`

[jpountz]

The track_scores allow users to tell Elasticsearch to compute scores of top hits and the maximum score when sorting by a field. This has a significant performance overhead and maintenance cost, yet I doubt that this is very useful: the score is only useful for ranking, so if you don't use it for ranking then it isn't useful?

[elasticmachine]

Pinging @elastic/es-search-aggs

[jpountz]

Discussed in FixitFriday: this change makes sense to us but we'd like to get more feedback from users before committing to it.

[ewstefan]

Our scenario that requires track_scores:

We make several attributes visible to users regarding an entity. Say 'size' and 'value'. Users search for entities via full text search. We provide a visual indicator as to how relevant each entity is based on the elasticsearch score.

Users can filter and/or sort on 'size' and 'value' and have their results displayed in a table or bubble chart. Seeing all attributes including relevance means : "Oh, while I see this entity has a bigger size or value than others, it may be more relevant though considering my search terms."

In our case this is useful information.

[coreation]

This might be ignorant on my part, but scoring is not just used for ranking, it's also used for filtering when you use the "min_score" in your query? I just stumbled on this issue because I want results back that have a minimum score, but of those results, I need to sort them by date.

[elasticsearchmachine]

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

[elasticsearchmachine]

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