Advancing African Accented Clinical Speech Recognition with Generative and Discriminative Multitask Supervision