Forgetful Fish Training Metrics

Self-play reinforcement-learning curves from the fishrl trainer plotted against training iteration.

N.B. The heuristic opponent was added to the training pool after iteration 9508.

N.B. Training scenarios and a forced loss when not taking a free attack were added after iteration 35592.

N.B. Run v2 is a fresh start (new model + corrected engine): the iteration counter resets and win rates re-climb from random.