RFR: Fix humongous object reclamation

[Roman Kennke]

When reclaiming humongous objects, we consider is_marked_current(),
however, we already swapped the bitmaps, and thus is_marked_prev()
would be correct.

It blew up on me once with SPECjvm2008 and couldn't reproduce it. I
tried for 1 hour to make a testcase, and it blew up once too, but not
reliable.

The fix seems rather obvious though, so here it is:
http://cr.openjdk.java.net/~rkennke/fixhumongousreclaim/webrev.00/

Ok to push?

Roman