[arch] move the atomic ops into a separate header

Now you need to include arch/atomic.h to get to the atomic routines.
This simplifies a recusion issue in the way arch/ops.h included
arch_ops. Also just generally makes things cleaner.

kernel/mp.c

@@ -8,6 +8,7 @@
 
 #include <kernel/mp.h>
 
+#include <arch/atomic.h>
 #include <arch/mp.h>
 #include <assert.h>
 #include <kernel/spinlock.h>