RFR: Enforce GC workers constraint

[Zhengyu Gu]

> I can imagine such a case. It is based on idea that we'd need to cut down the number of parallel
> threads to avoid scheduling overheads during pause where there is no need for too many threads. This
> is different from having lots of concurrent threads that share the work during longer concurrent phases.
> 
> Let me think about this for a while. Meanwhile, maybe you can estimate how much Shenandoah code goes
> away with this?

I guess we can not cut down those code, cause in concurrent phases, we 
still need to process all queues seeded by parallel phases.

I am curious about how can you estimate works for lesser workers. I am 
experimenting dynamic workers calculation algorithm - current sizing 
algorithm pretty much dictated heap size, e.g. 32M per GC thread, which 
only takes 256M to max out 8 workers on my laptop.

Thanks,

-Zhengyu

> 
> -Aleksey
> 
>