Consider using LuceneDictionary in suggester

Currently Lucene suggester is based on DocumentDictionary which builds suggestions upon stored values of a certain field (in this case :suggest), however it may be better to stick to plain indexed terms via a LuceneDictionary as this would allow to save some space in the index (:suggest field wouldn't have to be stored) and we can leverage per index (configurable) analyzer in order to tweak how suggestions will be returned: using a KeywordAnalyzer would result in same behaviour we currently have, using a tokenizing Analyzer will result in term level suggestions (tokens instead of field values).