Modernisation and consistency of regret and liap values

[tturocy]

As a bit of preparatory work towards #642 , this does a bit of refactoring:

• For mixed strategy profiles, be consistent in using the cached payoffs for both liap value and regret
• Refactor some common code (e.g. maximising a function over a container of values) into a suitable template
• Make mixed behaviour profiles parallel with mixed strategy profiles in the implementation style of liap value and regret.