Optimize mono_gc_bzero and mono_gc_memmove to closely match native performance.