Review TokenFilterFactory.getSynonymFilter() implementations

[romseygeek]

#33702 introduced a new method on the TokenFilterFactory interface allowing filters to return specialized versions of themselves for synonym parsing. Currently only the multiplexer implements this, to return the original token. We should review all the filter factories shipped with elasticsearch to see if any others need changing.

[] AsciiFoldingFilter -> should return only the folded token
[] CJKBigramFilter -> ignore if emitUnigrams = true
[] CommonGramsTokenFilter -> ignore
[] CompoundWordTokenFilterBase -> ??
[] EdgeNGramTokenFilter -> ??
[] Fingerprint & MinHash -> shouldn't be used with synonyms anyway...
[] Keyword repeat -> ignore
[] NGramTokenFilter -> ??
[] Shingle -> shouldn't output unigrams
[] SynonymGraph & Synonym -> should we allow multiple synonym chains?
[] Phonetic -> ignore
[] WordDelimiterGraph & WordDelimiter -> ??

[elasticmachine]

Pinging @elastic/es-search-aggs

[romseygeek]

Having thought about this some more, I think we should also ignore the ngram and shingle filters here, as synonyms generally are applied to whole tokens. Similarly, Synonym and SynonymGraph should also be ignored, because applying synonyms to a synonym file will make the end result very difficult to reason about.

I can see arguments both ways for WordDelimiterFilter and CompoundWordTokenFilterBase, but I think the simplest solution will be to ignore both for now. We can add a setting in future to allow them to be used when parsing synonyms if there's an obvious user need for it.