RFR: Move heuristics from ShCollectorPolicy to ShHeap

[Roman Kennke]

Hold on with review. I'm going to extend it to completely decouple
policy and heuristics.

Roman

> Another refactoring in preparation of interleaving GC. It moves the
> _heuristics out of ShCollectorPolicy and into ShHeap and (somewhat)
> decouples heuristics from ShCollectorPolicy.
> 
> The reasoning behind this is that for interleaving minor/major GC I plan
> to put a bunch of stuff into a 'context': control-thread, heuristics,
> workers, cset, fset, and probably some more. Each context will work
> somewhat independently on some part of the heap (e.g. young vs. old
> gen). There will be one heuristic per context, but there will only be
> one CollectorPolicy per GC. ShenandoahCollectorPolicy should eventually
> become a pure implementation for CollectorPolicy and nothing else. Also
> note that ShHeap is only a temporary home for the heuristics, they will
> eventually go to the context class. The point of this patch is to
> decouple heuristics from ShCollectorPolicy.
> 
> All the should* and can* methods are now called directly without hopping
> through ShCollectorPolicy. However, all the record* methods remain in
> the policy for now. I am not sure yet what to do with them: we may want
> to call them per-context (e.g. a context recording only its own GC
> activity) or globally (e.g. for stats) or both. I need to think this
> through when I do the multi-context stuff.
> 
> http://cr.openjdk.java.net/~rkennke/heuristics-refactoring/webrev.00/
> 
> Passes hotspot_gc_shenandoah
> 
> Ok?
> 
> Roman
>