Link, Transport, Integrate: A Bayesian Latent Class Mixture Modeling Framework For Scalable Algorithmic Dementia Classification In Population-Representative Studies