Signing a PDF file containing an image causes distortion to the image (using --clearsign).
gpg --clearsign --output='dices - signed with gpg.pdf' 'dices - unsigned.pdf'
--clearsign may only be used for plain text documents due to line ending conversion etc.
The surprising thing is that it works at all. I wouldn't be surprised if certain would simply reject it as "not a pdf" given that the "%PDF-1.x" marker isn't at the beginning.
PDF is a "text-based format" which then includes binary contents (e.g. compressed blocks) and uses byte offsets (so not really text).
For this specific file you could get away with the non-standard --not-dash-escaped option, as the issue on this is caused by 0a 2d being escaped into 0a 2d 20 2d:
gpg --clearsign --not-dash-escaped --output='dices - signed with gpg not dash escaped.pdf' 'dices - unsigned.pdf'
Even if you only used gpg to validate it, so that you can use --not-dash-escaped, certain pdf contents could confuse it, though. If you want to keep the signed pdf readable without gpg, the real solution would be to use a detached signature.