Piezo PIFO trees: More thoughtful data gen

[anshumanmohan]

I have adjusted the synthetic data generation script a little, when run with the --piezo flag.

Before:
It used to just randomly generate 100 commands, which were pushes or pops. This meant that overflow and underflow were totally possible. Even if you made the queue resistant to over/underflows (so that the show just went on until all 100 commands were done) we could end the program with packets still left in the queue. This was bad because:

• Failing by under/overflowing is actually quite cheap, so this compromised our "average cost per push" numbers, for example.
• Pushing something and then never popping it also leaves some questions when trying to find average costs.

After:
Now we generate commands with a little more care, making sure that we never have underflows or overflows, and that we always pop out everything we push in. We still generate 100 commands total. This alleviates the issues above.

[anshumanmohan]

Sorry that the tests appear to fail. They passed when I merged the branch, but my brief restoration of the branch has triggered a rerunning of the tests.