Optimize KPK bitbase propagation

[RainRat]

Summary

• build a predecessor graph while initializing the KPK positions and seed a frontier with known wins and draws
• propagate result changes through the frontier queue so only affected positions are reclassified while leaving the bitbase population unchanged
• add helper traversal on KPKPosition so classification reuses generated successors and re-queues only dependent predecessors when a state becomes known

Testing

• python3 test.py (fails: ModuleNotFoundError: No module named 'pyffish')

---

https://chatgpt.com/codex/tasks/task_e_68ce403bcd688330871ec58926999cd6