refining dict-typing in inference

added reward function normalization, hidden eval set

added canonical evaluation harness, unified DIME index, deterministic replay guarantees