RFR: Split live data management for allocations and GCs

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/liveness-split/webrev.01/

These are bits and pieces for allocation pacing work.

Pacer needs to disambiguate between live data produced by the allocations, and the live data
produced by GC marking work, in order to figure out marking progress. (Traversal needs it too, for
different reasons). It makes sense to just split the paths for allocations and GCs, and act
accordingly at each one.

ShenandoahAllocImplicitLive naturally goes into increase_live_data_alloc_words. Pacer would inject a
call into increase_live_data_gc_words.

Testing: hotspot_gc_shenandoah, pacing experiments

Thanks,
-Aleksey