The fuzzy query is pretty fast though it use Levenshtein distance,does it enumerate all terms?
no it uses a Levenshtein automaton that is build at runtime from the query term and efficiently intersected with the term dictionary. Note this only works up to LD 2