need help reviewing blocker patch for DERBY-1681 - Regression (wrong results): Join predicate can be ignored for left-most child in a chain of nested unions.

[Mike Matrigali]

I can act as committer on this one, will run a set of nightly tests. 
Would like at least one other pair of eyes.  Army says this one is a lot
more straight forward than 1633, I had thought that this was dependent
on 1633, but only the test is.  The fix is independent.
As usual with army's changes it is well documented.  From my review
the change does what the comments and the attached html says.