change example code to do mrb_close()

I was hoping this would cause valgrind to complain less, but there is still
a lot of memory leaked (does mrb_close() actually free all of the managed
blocks?)  Anyway this helps somewhat and is good practice

src/codegen.c

@@ -2291,6 +2291,7 @@ p(fib(30), \"\\n\")\n\
   codedump_all(mrb, n);
 #endif
   mrb_run(mrb, mrb_proc_new(mrb, mrb->irep[0]), mrb_nil_value());
+  mrb_close(mrb);
 
   return 0;
 }