I thought I had run all the implementations on my
HBASE-22386 patch, but I guess not. Some minor issues:
- There's a <= / < mixup in LocalTreeLockManager *LockBelow logic that's having it look 1 layer deeper than it needs to.
- We should org.junit.Assume that TestTreeLockManager's sync field isn't the null implementation, because we test internal details that the null implementation wouldn't even use.