Completion regex query uses UTF32 automaton [LUCENE-8480]

The completion regex query builds an UTF-32 automaton but the completion FST uses UTF-8 internally. This makes the matching of any non basic latin character impossible in a regex completion query.

---
Migrated from [LUCENE-8480](https://issues.apache.org/jira/browse/LUCENE-8480) by Jim Ferenczi (@jimczi), updated Oct 18 2018

Start by tracing the completion regex query and how it builds its automaton, then compare that encoding with the completion FST's UTF-8 representation. Reproduce the failure with a non-basic-Latin character and verify that regex completion matching works for such characters once the encoding mismatch is addressed.

Written by the indexing model from the issue text.