Fix test_stack failure if AO_PREFER_BUILTIN_ATOMICS (GCC/Aarch64)

Fix test_stack failure if AO_PREFER_BUILTIN_ATOMICS (GCC/Aarch64)

As of gcc-5.4/aarch64, not all of the built-in double-word atomics
are implemented correctly.

* src/atomic_ops/sysdeps/gcc/aarch64.h [!__clang__
|| AO_AARCH64_ASM_LOAD_STORE_CAS] (AO_double_load,
AO_double_load_acquire, AO_double_store, AO_double_store_release,
AO_double_compare_and_swap, AO_double_compare_and_swap_acquire,
AO_double_compare_and_swap_release, AO_double_compare_and_swap_full):
Define (asm-based implementation) even if AO_PREFER_BUILTIN_ATOMICS
or AO_THREAD_SANITIZER; update comment.