ARM64: the collector can miss a thread leaving a GC-free zone #142

When the collector already holds the state lock, it sets gPauseForCollect with a CAS and then reads each thread's mGCFreeZone with a plain load and no full fence. On weakly ordered CPUs (ARM64) the load can complete before the CAS is visible, so a thread can keep running during marking, or the collector can wait on a thread that entered a zone without signalling. x86 is not affected.