RFR: Fast-forward over humongous regions to keep "current" non-humongous

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/freeset-ff/webrev.01/

There is an interesting corner case that shows up intermittently in our testing:
ShFreeSet::current() fails with "should be humongous" assert. This is caused by the absence of
fast-forwarding over the humongous regions in allocate_contiguous: it may so happen that the "next"
region after current humongous allocation is humongous too, if heap was fragmented enough.

Testing: hotspot_gc_shenandoah, stressing the failing test

Thanks,
-Aleksey