Highlighter picks wrong offset for fragment boundaries [LUCENE-2587]

I have written a new Fragmenter since we need fragments for hitlines to be on sentence boundaries and not cross paragraphs.
When using it with org.apache.lucene.search.highlight.Highlighter, I get hitlines that starts with ". ", "? ", "! "...

Consider the text "A b c d e. F g h i j! K l m n o. "
which become the tokenstream : (A) (b) (c) (d) (e) (F) (g) (h) 🛈 (j) (K) (l) (m) 👎 (o)

If the fragmenter return isNewFragment() = true on F and K and Highlighter pick the middle fragment, lets say we search on "g" the hitline becomes:
". F <B>g</B> h i j"

The reason, it seems, is that the offset to the fragment boundaries found by taking the endOffset of the last token in a fragment ,
not the startOffset of the first.

TJ

---
Migrated from [LUCENE-2587](https://issues.apache.org/jira/browse/LUCENE-2587) by Terje Eggestad, 3 votes, updated Aug 15 2022
Environment:
```
Java 6 + Lucene 3.0.2
```

Attachments: [IMSentenceFragmenter.java](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-2587/IMSentenceFragmenter.java), [LUCENE-2587.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-2587/LUCENE-2587.patch), [TestIMSentenceFragmenter.java](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-2587/TestIMSentenceFragmenter.java)
Pull requests: https://github.com/apache/lucene/pull/1069

Read the attached IMSentenceFragmenter.java and TestIMSentenceFragmenter.java, then review LUCENE-2587.patch alongside org.apache.lucene.search.highlight.Highlighter. Use the reported sentence-boundary example as the reproduction; done means the selected hitline no longer begins with punctuation and the attached test demonstrates correct fragment offsets.

Written by the indexing model from the issue text.