Redo: Avoid stub when calling to write-barrier from C2, remove all related code

[Roman Kennke]

This is a more conservative version of the previous patch. It preserves
the exact code paths for interpreter and C1 versions of the stub,
including saving+restoring the vector regs. C2 is changed to call
straight into runtime as C2 obeys the C calling convention (albeit a
little conservatively, apparently disabling XMM spills around calls).

http://cr.openjdk.java.net/~rkennke/remove-c2-wb-stub/webrev.03/

Testing: tier3_gc_shenandoah, specjvm+scimark+aggressive which was
failing with previous patch

WDYT?

Roman