RFR: Improve/more detailed timing stats for Traversal GC

[Roman Kennke]

We've two problems in the timing stats in Traversal GC:
- the update-roots pass uses final_traversal_gc_work and thus overrides 
the previous final-traversal roots timings.
- finishing the queues is done in the same pass as root scanning, by the 
same workers.

This patch introduces a new set of worker timings for 
traversal-update-roots and a new worker timing item 'Finish Queues'.

http://cr.openjdk.java.net/~rkennke/traversal-stats/webrev.00/

example:

https://paste.fedoraproject.org/paste/bobw7OXRrLmDAg8vdiYAOQ

It shows that we're doing a *lot* of finish-queues stuff there. I 
suspect that this is where MWF will help a lot.

Ok to push?


Roman