Deprecate keyword_repeat in favor of the new multiplexer filter

[jpountz]

Looking at the docs, keyword_repeat seems to be mostly designed as a way to have both stemmed and original tokens indexed, which is now better handled by the multiplexer filter.

Should we deprecate keyword_repeat?

[elasticmachine]

Pinging @elastic/es-search-aggs

[cbuescher]

One caveat might be that the currently according to the docs "the Multiplexer does not work for Shingle or multi-word synonym token filters declared in the filters array because they read ahead internally which is unsupported by the multiplexer". I'm not entirely sure if this makes them a full substitute for the "keyword_repeat" filter of if that also has issues with multi-word synonyms.

[romseygeek]

The multiplexer essentially creates branches in the TokenStream and passes each token one-by-one to each branch in turn, so any filter within a branch that reads ahead in its tokenstream is going to break things. However, you can place filters after the multiplexer which in effect merge the stream again, and reading ahead will work fine there.

I think this is fine to do, we may well end up deprecating keyword_repeat in lucene as the conditional filter works as a more general replacement in any case.

[elasticsearchmachine]

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