don't count the tasks at each step. this is sufficiently expensive to mess up the results.

src/engine.c

@@ -2881,7 +2881,7 @@ void engine_step(struct engine *e) {
     submask |= 1 << task_subtype_tend;
   }
 
-  if (e->verbose) engine_print_task_counts(e);
+  // if (e->verbose) engine_print_task_counts(e);
 
   /* Send off the runners. */
   TIMER_TIC;