benchmark strcpy_s/memcpy_s overhead 90%

there is no measurable --disable-constraint-handler speedup,
but a huge _s slowdown of ~90%.
The size overhead for the constraint-handler is 35%-100%.

Use proper timings, asm rdtsc