Capability-Routed Guard: Defending Large Reasoning Models Against Reasoning-Centric Jailbreaks | AIChainDay