[Java] MurmurHasher doesn't take the final bytes into account if size % 4 != 0

Minimal test case: all buffers of size 1 currently hash to the same value (-1684229222)

To my (admittedly untrained) eye, it looks like should range from `length - 1` rather than `index - 1`?

Also, should the nearby `index += 1` be removed, so that `i` and `index` don't meet in the middle? (`index` is being checked in the loop condition)

**Reporter**: [James Henderson](https://issues.apache.org/jira/browse/ARROW-15175) / @jarohen

_{**Note**: *This issue was originally created as [ARROW-15175](https://issues.apache.org/jira/browse/ARROW-15175). Please see the [migration documentation](https://github.com/apache/arrow/issues/14542) for further details.*}

Open java/memory/memory-core/src/main/java/org/apache/arrow/memory/util/hash/MurmurHasher.java at the referenced line and reproduce the issue with buffers whose size is not divisible by four, especially size 1. Trace the loop bounds and final-byte handling, then verify that buffers with different final bytes no longer produce the same hash.

Written by the indexing model from the issue text.