RFR: AArch64 fixes to C1 SFX barriers

[Roman Kennke]

More testing on aarch64 revealed some more bugs/missing stuff in C1 SFX
barriers.

- A whole bunch of stuff was missing to actually call the SFX entry
points with the address. We probably got away with that by accident.
- leal on aarch64 wants a double-register, where x86 wants a
single-register. I special cased that part in ensure_in_register().
Alternatively we could make special platform-specific implementations in
shenandoahBarrierSetC1_XXX.cpp but that is probably overkill.

http://cr.openjdk.java.net/~rkennke/c1-sfx-aarch64/webrev.00/

Testing: hotspot_gc_shenandoah on aarch64

Ok?