Fix infoset realisation probability for absent-minded information sets

[d-kad]

Description

This PR fixes information set realisation probability for games with absent-mindedness. The previous implementation summed node realisation probabilities across all members, double-counting it for the nodes along the same path, with probabilities summing above 1 in this case.

The corrected infoset realisation probability is the probability the information set is visited at least once. The implementation identifies reentry nodes (members whose path from the root passes through the same information set) and excludes them from the sum.

C++ changes:

• Add m_absentMindedReentries cache in GameTreeRep, populated inside BuildOwnPriorActions
• Add GetAbsentMindedReentries
• Add m_infosetProbs cache, populated inside ComputeRealizationProbs

Test changes:

• Add test_absent_minded_infoset_prob covering noPR-AM-driver-one-player.efg, noPR-action-AM.efg, and noPR-action-AM-three-chain.efg with uniform probability distributions over actions

Closes #826

[tturocy]

The only thing that is giving me pause here is whether the need to run the unreachable nodes check and then the way the absent-minded nodes check is being done is too inefficient.

Wouldn't it be simpler to do the sums and then subtract off the probabilities from the re-entries? That way if the set of re-entries is empty that would be fast.

[tturocy]

We've realised with further discussion that there are some additional adjacent considerations to defining some other quantities that should be handled as part of these changes (as they interact with the definition of information set realisation probability) - see discussion to be added to #826

[tturocy]

@d-kad I am having difficulty correctly resolving the merge conflict in the tests - as you've been working on these I would hope you would be able to see the correct solution more quickly than I!