Reasoning Model Is Stubborn: Instruction Overriding in Reasoning Models