RFR: Store checks should run most of the time

[Aleksey Shipilev]

http://cr.openjdk.java.net/~shade/shenandoah/store-check-cm-ur/webrev.02/

Current store check code is wrong at least for x86: it checks ShenandoahUpdateRefsEarly which is
ccstr coerced to the bool. Also, it has to actually check for both CM and UR flags, not choosing one
statically. After Roman's change that makes storeval barriers unconditional, we can actually only
check for EVAC in progress, and thus cover most of the stores most of the time.

Did AArch64 change blindly.

Testing: hotspot_gc_shenandoah, simple jcstress with -XX:+ShenandoahStoreCheck

Thanks,
-Aleksey