Consolidate validation for 'docvalue_fields'.

[jtibshirani]

This improves modularity and also fixes some issues when docvalue_fields is
used within inner_hits or the top_hits agg:

• We previously didn't resolve wildcards in field names.
• We also forgot to enforce the limit index.max_docvalue_fields_search.

[elasticmachine]

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

[cbuescher]

LGTM

[jtibshirani]

Thanks @cbuescher for the review. This is technically breaking, since we now enforce the index.max_docvalue_fields_search limit in places where we didn't used to. What's your thought on whether this should be backported or not? I will push a note to the breaking changes docs before merging.

[cbuescher]

we now enforce the index.max_docvalue_fields_search limit in places where we didn't used to.

As far as I understand the PR this would be limited to using "doc_values" in Top Hits Aggregation and inner hits queries? And the now enforced 100 field limit would probably only be hit when using some sort of wildcard there and there are a huge number of fields?
If there are that many fields, chances are the user needed to update the index.max_docvalue_fields_search already für regular searches where the limit already applies, in which case the limit might already be high enough. And the fact that the limit is changeable dynamically on all indices should make solving this issue in case you run into it also a manageable task.

As far as I'm concerned this should already have triggered an error (I think we fact that we didn't check the limit until now can be considered a bug), so I wouldn't be opposed to backporting to 7.9, maybe adding a note to the changes doc for the cautious. I did a bit of searching in the forums and other support resources where users so far ran into the existing limit, one was in SQL but should be fixed (#44062), and the main problem seems to be this open Kibana issue (elastic/kibana#22897), but that one is concerned with the limit we already check. What's your thoughts on this?

[jtibshirani]

And the now enforced 100 field limit would probably only be hit when using some sort of wildcard there and there are a huge number of fields?

This is a good point, it's unlikely that a user would have exceeded the limit if wildcards weren't even supported. I also feel like it makes sense to treat this as a 'bug' and backport to 7.9. (Update: this actually turned out to be 7.10).