fix: update kuhnpoker env

[Benjamin-eecs]

This pull request focuses on improving the logic and robustness of the Kuhn Poker environment in textarena/envs/KuhnPoker/env.py. The changes address several issues, including incorrect game state handling, winner determination, and round transitions, while also adding safeguards to prevent errors during gameplay.

Game state and logic fixes:

• Corrected the initialization of current_round in the reset method to start from 0 instead of 1.
• Fixed the winner determination logic in _init_round to compare the correct chip counts for each player.
• Added a check in _init_round to return immediately after setting the game as a draw or declaring a winner, avoiding further processing.

Gameplay flow improvements:

• Added a check in the step method to immediately return if the game is already marked as done, ensuring no further actions are processed.
• Updated the step method to prevent calling self.state.step() after round-ending actions, avoiding player-switching issues.

Round and game-ending logic:

• Enhanced _set_round_winner to directly end the game and declare a winner if the maximum number of rounds is reached, instead of waiting for another action.
• Ensured that non-final rounds start immediately after announcing the round result, improving gameplay continuity.

Miscellaneous:

• Updated _handle_showdown to use ta.GAME_ID as the sender ID for observations, ensuring consistency in game messages.

[LeonGuertler]

Looks great! Two minor points:

1. Do you think it would be possible to use the manually_update_current_player function instead of straight updating the attribute? In the state class, for example, player updates are prevented on invalid action, I worry the proposed changes my clash with that.

2. Also, do you think we can solve the check for round completion differently than overwriting the get_observation function? Just to make sure future changes won't affect this unnecessarily.