Implement Self-playing Adversarial Language Game Enhances LLM Reasoning

[cstorm125]

I'd like to add Self-playing Adversarial Language Game (SPAG) as an algorithm.
Will test on one of textarena's 2-player, zero-sum language game.

paper: https://arxiv.org/html/2404.10642v1#S2
github: https://github.com/Linear95/SPAG