Add tests for reproducibility of stochastic results

[drvinceknight]

I'd like to add a property based test similar to https://github.com/Nikoleta-v3/Axelrod/blob/635/axelrod/tests/integration/test_matches.py but that checks that when setting a seed we get the same results for stochastic strategies (those tests there only check for 'well behaved deterministic' strategies. I have a hunch that this isn't actually true as we're not setting numpy's seed...

If that's the case and both seeds need to be set, we should document how to get reproducible results and either:

1. Indicate that you should set both seeds (numpy and stdlib);
2. Write a little helper function that does this (and just document it: axelrod.seed() or something like that...