I think it is important for SDC using topup to have different phase encoding directions, if both images have actually been acquired in that way of course. But as it was mentioned earlier in this thread, the b0 images have also very different intensity between the two acquisitions, perhaps due to the fact that the acquisition parameters are different between the two. For instance, the RepetitionTime for the dwi image is 2.96251 s while it is 5.9 s for the fmap image. This difference of intensity may not help topup either.
It seems like a good approach in that case to switch to the fieldmap-less options mentioned by @Steven and that you are already testing:
For synB0-DISCO, if you wanted to integrate it into the qsiprep workflow, you could be inspired by this discussion:
or by a similar approach for bold images, synBOLD-DISCO and fmriprep, it is basically the same idea: