Avoid race condition in crutest rand-read/write

[mkeeter]

Previously, the IO tasks could stop before the stat-gathering main loop, so you could get nonsensical readings like

 713.4 MiB/sec ± 14.2 MiB/sec      (713 IOPS)
 683.6 MiB/sec ± 11.2 MiB/sec      (683 IOPS)
 698.8 MiB/sec ± 409.6 KiB/sec     (698 IOPS)
 722.2 MiB/sec ± 10.2 MiB/sec      (722 IOPS)
 674.4 MiB/sec ± 13.6 MiB/sec      (674 IOPS)
   737 MiB/sec ± 5.8 MiB/sec       (737 IOPS)
 709.6 MiB/sec ± 16.8 MiB/sec      (709 IOPS)
 728.6 MiB/sec ± 10.6 MiB/sec      (728 IOPS)
 723.6 MiB/sec ± 15.2 MiB/sec      (723 IOPS)
 694.6 MiB/sec ± 8.6 MiB/sec       (694 IOPS)
 713.2 MiB/sec ± 7.2 MiB/sec       (713 IOPS)
 stopping guest [0 jobs remaining]
----------------------------------------------
average read performance: 650.1 MiB/sec ± 196.6 MiB/sec (650 IOPS)

Notice the average read performance is nowhere near the average values reported during runtime!

This is because the last two IO samples are captured as the system is shutting down. Here are the raw samples:

[733164339.2, 762943897.6, 728550604.8, 705062502.4, 732325478.4, 733164339.2,
 746586112.0, 767977062.4, 692899020.8, 721420288.0, 778882252.8, 766718771.2,
 726453452.8, 761685606.4, 775107379.2, 752877568.0, 742811238.4, 774687948.8,
 719323136.0, 737358643.2, 755394150.4, 740294656.0, 3774873.6, 0.0]