Fix netstats rollover

[graeme-a-stewart]

Change the way that network statistics are calculated so that if any stat rolls over (64bit uint overflow or another issue with the network counters), such that during an iteration

current_value < previous_value

Then the change is ignored for this cycle and a warning is printed.

Fixed a bug in the network stats test scripts (target values not updated correctly if --blocks was specified).

Fixed a few minor flake8 issues in test scripts.

Closes #180

[amete]

Thanks a lot @graeme-a-stewart. The updates look good to me. Instead of doing now-start, we now do now-last and increment the counters accordingly, except when there is a drop.

The only bit I think can be improved, if we want to monitor how often the check fails, is to keep an internal counter, say network_error_counters, increment this when we bump into a decrease (in addition to the error message), and push this into the JSON at the end of the job. The error message itself will probably be fairly hard to fish out in production jobs.

Having said that, I don't really feel too strongly on that. Therefore, let me approve as is and let you decide 😄

[graeme-a-stewart]

Thanks for taking a look @amete. It would be interesting to think about error counters, but I would prefer that we added it more generally to the Imonitor class and that all of the monitors could use it. Otherwise we get into a situation where we're handling a monitor in a special way, which compromises the design of the code.

So my inclination is to accept this PR and create a new issue if we want to handle errors in a more systematic way (also, there was my thought to introduce more formal logging in general, #106).