4D ASR: Joint modeling of CTC, Attention, Transducer, and Mask-Predict decoders