TestCollationDocValuesField.testRanges failure [LUCENE-6206]

This test failed in branch_5x:

```
ant test -Dtestcase=TestCollationDocValuesField -Dtests.method=testRanges -Dtests.seed=8729A11DAE9AE179 -Dtests.multiplier=3 -Dtests.slow=true -Dtests.locale=lt -Dtests.timezone=Etc/GMT-8 -Dtests.asserts=true -Dtests.file.encoding=UTF-8
```

The problem is, with lithuanian collator, the generated collationkey does not agree with collation order.

In other words: collator.compare("yi", "iy") != collator.getCollationKey("yi").compareTo(collator.getCollationKey("iy"))

Simple test:

```Java
import java.text.Collator;
import java.util.Locale;

public class TestLithuanian {
public static void main(String args[]) throws Exception {
Collator collator = Collator.getInstance(new Locale("lt"));
int value = collator.compare("yi", "iy");
System.out.println(value); // returns 1
int value2 = collator.getCollationKey("yi").compareTo(collator.getCollationKey("iy"));
System.out.println(value2); // returns -1
}
}
```

Note that if you change java.text.Collator to com.ibm.icu.text.Collator, they are consistent. Looks like a bug in JDK collation.

---
Migrated from [LUCENE-6206](https://issues.apache.org/jira/browse/LUCENE-6206) by Robert Muir (@rmuir), updated Aug 31 2015

Start by running the reported ant command for TestCollationDocValuesField.testRanges with the Lithuanian locale and seed. Read the test and compare java.text.Collator ordering with its CollationKey ordering, then check the noted com.ibm.icu.text.Collator behavior. Done requires a confirmed explanation or a project change that prevents this failure.

Written by the indexing model from the issue text.