Contrastive learning of strong-mixing continuous-time stochastic processes