Upload verify.py with huggingface_hub

Verifier: add the agent-mode taxonomy, the planning-is-toolless checks and the intervention table

Add the measurement corpus: 8,199 runs, 110,711 ledger events, 14,008 refusals, with scorer and verifier