Fix AO_nop_full and AO_load_read for AArch64

Fix AO_nop_full and AO_load_read for AArch64

* src/atomic_ops/sysdeps/gcc/aarch64.h: Place include of
read_ordered.h to the end of file (to be after AO_load definition).
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_nop_full): Force compiler
barrier for the uni-processor case.
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_load_acquire): Comment out
and add FIXME (due to read_ordered.h).
* src/atomic_ops/sysdeps/gcc/aarch64.h (AO_test_and_set,
AO_test_and_set_acquire, AO_test_and_set_release,
AO_test_and_set_full): Cast result to AO_TS_VAL_t (since
__atomic_test_and_set returns AO_TS_t value).