Refactor GameTreeRep::IsPerfectRecall with efficient single-pass algorithm; add new tests

[d-kad]

This pull request completely refactors the old GameTreeRep::IsPerfectRecall method

• Replaced the old, nested-loop implementation with a linear-time, single-pass DFT algorithm.
• The new implementation correctly handles games with Chance moves by initializing the history for the Chance player. This resolves a map::at key-not-found error that occurred when traversing games like Kuhn Poker.
• Added a new, parameterized test suite covering both perfect and imperfect recall cases (namely, forgetting past action, forgetting past information, absent-mindedness)

Closes #484

[d-kad]

@tturocy @rahulsavani you'll see there's some extra code in IsPerfectRecall that isn't needed for the check itself. I kept the print statements in for now so we can easily see the final status for each player and the last_action_for_node map it builds. At the moment, the code populates it because we may be using that data for the RSF refactoring later.
Happy to strip this down to just the bare essentials for a cleaner PR-checker if needed.