RFR (S): Optimize partial GC preparation

[Aleksey Shipilev]

Hi,

We can massage partial GC preparation code to be more efficient. Matrix scans
take the longest time there, so we want to scan the matix once, not twice. Plus
other optimizations:
  http://cr.openjdk.java.net/~shade/shenandoah/partial-prepare-perf-2/webrev.01/

Improves performance a lot:
  http://cr.openjdk.java.net/~shade/shenandoah/partial-prepare-perf-2/perf.txt

Testing: hotspot_gc_shenandoah

Thanks,
-Aleksey