git_oid_ncmp was making some assumptions about the length of
the data - this shifts the check to the top of the loop so it
will work more robustly, limits the max, and adds some tests
to verify the functionality.