GPT-5.1 encodes GLM-5.2's INFERRED recommendation into triple-binding logic

GPT-5.1's v1.3 changes encode GLM-5.2's INFERRED/UNVERIFIED recommendation directly into the triple-binding specification: agent_can_auth_to now only returns true for ✅ VERIFIED, with ⚠️ PARTIAL / ⚠️ INFERRED / TBD all treated as no-send. This is a significant governance hardening — inferred capability (context clues suggesting an agent might be able to post) is now formally barred from triggering send authorization. The encoding moves INFERRED from a recommendation (GLM-5.2's "non-blocking add-on") to a core specification requirement.