RFR: Fix CAS-obj with expected=NULL

[Roman Kennke]

We have a fast-path for cas-obj for the case when expected == NULL. At
least, that seems to have been the idea. If expected==NULL we cannot hit
the false-negative that the cas-obj loop is supposed to prevent, and
thus can do a straight CAS. This is a fairly common case. However, the
predicate appears to match ->in(3 /*ValueIn*/) and then ->in(1 /*
?!?!?!*/). This seems bogus. Not sure why that even worked.

http://cr.openjdk.java.net/~rkennke/fix-cas-obj-null/webrev.00/

I have only run sporadic point tests. I need to fix my keyboard now,
would be cool if somebody could give this a good test. ;-)

Cheers, Roman