According to @apunjani, true per-frame reconstructions are only a cryosparc-tools script away because all the infrastructure for it already exists (it is simply not exposed as a specific job type for this use). I haven’t been able to try it yet because my institutional instance hasn’t been upgraded to v5 yet (I need the new login method because the person running the instance prefers not sharing the license ID with users).
For the zero-dose extrapolation step, the scripts in Chris Russo’s paper do work but take a very long time to complete, and are not straightforward to set up: you have to manually keep track of all files, etc. I ran them successfully on a dose series made with relion_movie_reconstruct (also very slow, memory hungry, and pretty “low-level” in terms of file management). I know Chris Russo’s group is working on a faster and more user-friendly extrapolation program (they had a poster about it at ccpem both last year and this year), but I don’t know when it will be publicly available.
I was also hoping to convince Ali to add both of these features (per-frame reconstructions and zero-dose extrapolation) to CryoSPARC, so it’s nice that we are at least two asking for it. 
It will be convenient to have this built-in, and like you I am curious to see what this unlocks for radiation-sensitive systems (we were recently chasing a regulatory disulfide and learned the hard way that it’s difficult…).