RFR: Process remaining SATB buffers in final mark/traverse loop instead of separate phase

[Roman Kennke]

We don't need an extra phase to process remaining complete buffer in
final-mark/-traverse, and we don't need all the loop versions twice with
and without SATB. We can just as well do that in one single loop, pretty
much the same as conc-mark/-traversal. The important part is to process
*in*complete thread-local buffers. We already piggy-back this on
final-thread-scanning in traversal, and we keep a small extra phase for
that in conc-mark. At the very least, this should keep the overall
generated code smaller.

Aleksey, can you give this some smoke performance runs to check for
possible regressions or maybe improvements?

http://cr.openjdk.java.net/~rkennke/satb-final-phase/webrev.00/

Thanks,
Roman