HunspellStemmer dedup not fully functional [LUCENE-3670]

public List<Stem> uniqueStems(char word[], int length) assumes that word.length is always same as length and in cases where it is not, duplicate stems are returned.
Quick patch may be applied to trim extra characters:

if (word.length > length) {
terms.add(Arrays.copyOf(word, length));
} else {
terms.add(word);
}

---
Migrated from [LUCENE-3670](https://issues.apache.org/jira/browse/LUCENE-3670) by Emir Arnautovic

Locate HunspellStemmer.uniqueStems(char[], int) in the Java source and inspect how the supplied length is used when collecting stems. Reproduce the case where the array is longer than length, then verify the change prevents duplicate stems while preserving the requested word length.

Written by the indexing model from the issue text.