[Bug]: Core dump in GetAgentMaxRegret() for 0-dimension / degenerate games

[AndresFerCervell]

Where did you find this bug?

Gambit CLI

What operating system are you using?

Linux

What version of Gambit are you using?

16.6.0

What happened?

Testing the guard proposed in logit to prevent a core dump for 0-dimension games such as "chance_root_5_moves_no_nonterm_player_nodes.efg", (#883) I proposed to add this code in test_nash.py:

pytest.param(
        EquilibriumTestCase(
            factory=functools.partial(games.read_from_file, "chance_root_5_moves_no_nonterm_player_nodes.efg"),
            solver=gbt.nash.logit_solve,
            expected=[
                [[]]  # Zero-dimension edge case (two players)
            ],
            regret_tol=TOL_LARGE,
            prob_tol=TOL_LARGE,
        ),
        marks=pytest.mark.nash_logit_behavior,
        id="test_logit_behavior_degenerate",
    ), 

right after the create_stripped_down_poker_efg test. The next error occurred:

Accessing agent_max_regret causes an exception when trying to reach the position 0 of an empty array, when the regret of a choiceless game should be 0.0.
The exception is:

IndexError: vector::_M_range_check: __n (which is 0) >= this->size() (which is 0)