RFR: Handle scenario that not be able to satisfy the request for active workers

[Aleksey Shipilev]

On 10/03/2018 11:09 PM, Zhengyu Gu wrote:
> Newly updated EvilSyncBug test reveals that it is possible workers->update_active_workers() might
> not be able to honor the request, e.g. it failed to create additional workers, due to memory or os
> limitations, etc.
> 
> Under those scenarios, we may get assertion failure (debug) or reserve wrong number of queues for
> workers (product), which could result hard to find errors downstream.
> 
> Webrev: http://cr.openjdk.java.net/~zgu/shenandoah/worker_scope/webrev.00/

*) "must be able to re_S_tore"

  70   assert(nworkers == _old_workers, "Must be able to retore");

Otherwise good!

-Aleksey