When calculating unique for chunked DictionaryArrays we currently run through all chunks and unify their dictionaries and then collect chunk indices. We could avoid the dictionary unification by using a generic hash.

See discussion here and here

Reporter: Rok Mihevc / @rok

Related issues:

• [C++] Query engine umbrella issue (is a child of)

_{Note: This issue was originally created as ARROW-12301. Please see the migration documentation for further details.}