Implement action probabilities

[tturocy]

At an information set belonging to the chance player, actions have assigned probabilities. This issue discusses exposing the interface to setting and getting these probabilities.

In C++, these operations are done using operations on GameInfosetRep, namely, GetActionProb and SetActionProb.

In Python, we propose to implement a property of Action, tentatively named 'prob', which will serve to get and set these probabilities.

As actions are numbers, they follow the usual convention in libgambit for dealing with numbers (the same as we use for representing payoffs). Therefore, 'prob' should be settable via a fractions.Fraction or decimal.Decimal, and should return the same, appropriate to whichever representation is used in the game, just as with payoffs.

Note that there is no constraint in libgambit that probabilities need to sum to one at any given time.

Relevant entries should be added to the documentation, and entries made in the test suite to exercise this property.