Inefficient FVhighlighting when set many HighlightedField. [LUCENE-7397]

when highlighting, search result org.apache.lucene.search.vectorhighlight.FastVectorHighlighter.java
getBestFragment method \~ FieldTermStack.java read whole doc's termvector every highlighted field.
It causes slow query when many highlight field

---
Migrated from [LUCENE-7397](https://issues.apache.org/jira/browse/LUCENE-7397) by donghyun Kim, updated Jul 29 2016
Environment:
```
CentOS release 6.4 (Final)
quad core 1.87
8gb memory
tested Elasticsearch - 1.5 with lucene 4.10.4
But i see mirrored Master version in github
https://github.com/apache/lucene-solr
```

Start in org.apache.lucene.search.vectorhighlight.FastVectorHighlighter.java at getBestFragment, then inspect FieldTermStack.java to trace how each highlighted field reads the document term vector. Reproduce highlighting across many fields and measure the repeated reads; the work is done when the unnecessary repeated term-vector processing is eliminated and the slow-query behavior is verified.

Written by the indexing model from the issue text.