RFR: Rework GC degradation on allocation failure

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/alloc-failure-rework/webrev.01/

I was doing the ClassUnloading fixes and realized that one of my stress tests did not fail as
expected with OOMe. Instead, Shenandoah tried futively to do the GC cycles over and over again,
never stopping. The fixes for that path are in this patch. It required some work in defining what
the GC progress is. New tests verify Shenandoah fails as expected.

Testing: hotspot_gc_shenandoah

Thanks,
-Aleksey