UnifiedHighlighter: Convert PhraseHelper to use SpanCollector API [LUCENE-7578]

The PhraseHelper of the UnifiedHighlighter currently collects position-spans per SpanQuery (and it knows which terms are in which SpanQuery), and then it filters PostingsEnum based on that. It's similar to how the original Highlighter WSTE works. The main problem with this approach is that it can be inaccurate for some nested span queries – #3363, #6518 (has the clearest example), #7854. Non-nested SpanQueries (e.g. that which is converted from a PhraseQuery or MultiPhraseQuery) are _not_ a problem.

---
Migrated from [LUCENE-7578](https://issues.apache.org/jira/browse/LUCENE-7578) by David Smiley (@dsmiley), 1 vote

Start by locating PhraseHelper in the UnifiedHighlighter and reading the SpanCollector API, then review the nested SpanQuery cases described in issues #3363, #6518, and #7854. Done means PhraseHelper uses SpanCollector while preserving accurate handling of nested span queries and existing non-nested phrase queries.

Written by the indexing model from the issue text.