NGramTokenizer strips whitespace, with no option to keep leading and trailing whitespace [LUCENE-3979]

org.apache.lucene.analysis.ngram.NGramTokenizer removes whitespace, making a search for literal strings like " test" and "test " equivalent to "test". Searching with relevant whitespace is sometimes desired, particularly where ngrams are used.

This could be fixed by either removing .trim() from the line shown below, or by providing a flag to specifically set trimming behaviour (keeping trim=true as the default so that existing code using this analyzer is not broken).

111: inStr = new String(chars).trim(); // remove any trailing empty strings

---
Migrated from [LUCENE-3979](https://issues.apache.org/jira/browse/LUCENE-3979) by David Mason
Environment:
```
n/a
```

Start in org.apache.lucene.analysis.ngram.NGramTokenizer at the line 111 call to String(chars).trim(). Compare the requested preservation of leading and trailing whitespace with the stated need to keep trim=true by default for compatibility. Done means literal strings such as " test" and "test " are no longer treated as equivalent to "test", without breaking existing default behavior.

Written by the indexing model from the issue text.