RFR: Out-of-TLAB allocation should overwrite stale copies

[Aleksey Shipilev]

One of the failures we were chasing for a while now was apparently caused by the
trouble with out-of-TLAB allocation during evacuation. Then, we could leave the
"alive" object with garbled ptrs, on which we'd then crash during subsequent
pauses. See more details in the patch comments.

Fix:
  http://cr.openjdk.java.net/~shade/shenandoah/evac-non-tlab/webrev.02

I also cleaned up evacuate_object while we are at it.

Testing: hotspot_gc_shenandoah (includes tests that were failing before)

Thanks,
-Aleksey