The Verification-Value Paradox: A Normative Critique of Gen AI in Legal Practice