ARM/v7-a: Force -O2 compilation of idle thread

This fixes a problem with -O0 SMP builds on ARM visible as kernel
data aborts whenever the idle thread executes.

The idle thread receives no stack pointer. At -O2, this is fine
as the wfi() call is inlined and stack operations in idle_thread
are optimized out. At -O0, the stack operations remain and wfi()
is not inlined, resulting in stack accesses in the idle thread
that cause data aborts.

Forcing -O2 behaviour was deemed the simplest solution for now.
Giving the idle thread a stack would have had larger verification
ramifications for what is now a fairly uncommon use case.

include/util.h

@@ -43,6 +43,7 @@
 #define UNUSED       __attribute__((unused))
 #define USED         __attribute__((used))
 #define FASTCALL     __attribute__((fastcall))
+#define FORCE_O2     __attribute__((optimize("O2")))
 /** MODIFIES: */
 void __builtin_unreachable(void);
 #define UNREACHABLE()  __builtin_unreachable()