RFR: Re-fix memory managers and memory pools usage and pause reporting

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/mx-pause-cycle/webrev.01/

We recently did the fix in MXBeans that made it report GC pauses only. Unfortunately, this wrecks up
notifications about GC heap usage, because our complete cycle has concurrent phases that are
responsible for reclaiming the memory. We cannot pass this information with GC-pause events only.

This change rewires our MXBeans handling: instead of "minor"/"major" managers that make little sense
in Shenandoah context, we have two separate managers, one reporting statistics for the entire GC
cycle ("Shenandoah Cycles"), and another is reports data specific to GC pauses ("Shenandoah
Pauses"). New test validate these managers report sensible data.

This patch partially reverts the work Zhengyu did last week.

Testing: hotspot_gc_shenandoah, new tests, eyeballing MXBeans notifications

Thanks,
-Aleksey