Add positive_score_impact to rank_features type

[mayya-sharipova]

rank_features field type misses positive_score_impact parameter
that rank_feature type has. This adds this parameter.

Closes #68619

[elasticmachine]

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

[mayya-sharipova]

@elasticmachine run elasticsearch-ci/bwc

[cbuescher]

Looks pretty similar to what rank_featuredoes, nice to get some insight how this is working. LGTM

[jpountz]

I had left it out on purpose on the first release of rank_features because my mental model is that rank_features is about sparse features. In that situation, positive_score_impact:false doesn't work well because it would contribute a higher score to a document that has the feature than to a document that doesn't have the feature, which feels a bit counter-intuitive? The example works around this problem by setting the same features on all documents but if this is something that is possible, I wonder if users wouldn't naturally go to rank_feature instead?

[mayya-sharipova]

@jpountz Thank you for the feedback, it makes sense, but it looks to me this should be clarified in the documentation.

by setting the same features on all documents but if this is something that is possible, I wonder if users wouldn't naturally go to rank_feature instead

But for the rank_feature field type as well, we don't enforce that all documents have this field. And similarly, if this field is used in the rank_feature query as a part of should bool query, a doc that has this feature will have a higher score than a doc that doesn't have.

Or are we implicitly expecting that all docs must have rank_feature field?

[jpountz]

+1 to clarify the documentation and recommend that rank_feature fields have a value for every document that matches the query when positive_score_impact is set to false.