CPU affinity-setting code should be cleaned up

[Rod Smith]

Although I've not tested it in start-iperf3, the following bug in Checkbox's network.py almost certainly applies to start-iperf3, too:

https://github.com/canonical/checkbox/issues/570

The affinity-setting code recently added to network.py comes from start-iperf3 with minimal changes, so that bug almost certainly exists in start-iperf3, too. We presumably haven't run into it in start-iperf3 because that script is usually run on servers that do provide NUMA information. They're also more powerful, so a bump up in CPU load might not cause problems. That said, this bug should be squashed, along with its twin in network.py.