Yep, as I suspected — it is indeed that the Q-scores are all uniform (presumably in all sequences, not just the first few).
This is impossible — it indicates that the authors of the study most likely created artificial Q scores for the reads, possibly because the original quality scores were lost. You could contact them for the true Q scores/explanation.
So these data will not work with dada2. You could use deblur (which does not use Q scores) for denoising.
Good luck!