Refactor MixedBehaviorProfile caching

[tturocy]

This re-writes the cached quantity calculations for MixedBehaviorProfile:

• Uses the preorder and postorder traversal of nodes provided by the game class;
• Thereby avoids recursion to descend the tree
• Cleanly separates the computation of each vector of quantities (which will be a help for further optimisation in future when we look at the data structures used to represent them)
• Because there is a dependency order in the cached quantities, implements a slightly more sophisticated cache to compute only what is truly needed.

[tturocy]

@rahulsavani I've completed my first go at rationalising the way we're doing caching in mixed behaviour profiles. I've still got a bit more to do, but this is advanced enough it is worth you having a look. All tests continue to pass, and also I checked a few of my standard extensive games in liap/logit and didn't notice any differences. Note that this is based on current master so it does not reflect yet the nomenclature changes for agent.

@d-kad I wanted to call your attention to this because I've implemented here a more sophisticated Cache object. This is something we should develop for the extensive game as well to manage all the information being computed there, especially your work. For the moment no action is required (in fact don't let it distract you from the thrust of your work) but we will circle around and refactor so we can track more carefully what's been computed.

And for both - note this new implementation uses both the preorder and postorder traversal versions of GetNodes() - and insofar as all tests pass it's further confirmation both now appear to be correct!