Classic query parser should allow autoGeneratePhraseQueries=true when splitOnWhitespace=false [LUCENE-7799]

#8584 disabled the option combination `splitOnWhitespace=false`/`autoGeneratePhraseQueries=true` because of how `autoGeneratePhraseQueries=true` is handled: a query chunk is treated as if it were literally quoted. When `splitOnWhitespace=false`, a query chunk can be multiple whitespace-separated words, and auto-quoting multiple terms will produce inappropriate phrase queries.

I have an idea about how to fix this: `autoGeneratePhraseQueries=true` is supposed to cause phrase queries to be constructed when multiple analyzed terms result from a single query word, e.g. when WordDelimiter(Graph)Filter splits words up. Maybe this could be re-implemented in terms of offsets, since all terms from the same original term share the same offsets.

---
Migrated from [LUCENE-7799](https://issues.apache.org/jira/browse/LUCENE-7799) by Steven Rowe (@sarowe), 1 vote, updated Apr 24 2017

Start by locating the Classic query parser handling of splitOnWhitespace and autoGeneratePhraseQueries, then review the change from #8584 and the offset behavior described here. Determine how terms from one original query word should form phrase queries without treating multiple whitespace-separated words as a phrase. Done means the option combination works correctly and existing parser behavior remains intact.

Written by the indexing model from the issue text.