Hi Matjaz,
I'm not on the team that's responsible for this now (that's Learner), but I've been in conversations about this feature in the past and I just touched base with Jasper, who is the edX product owner for this.
EdX has been adopting a more experiment-focused approach to development, with faster prototyping of features to assess impact and demand before committing the resources for a large project. The PR you referenced was a part of that. So we don't have an official commitment at this point to go forward with this feature, or exactly what shape it will take. There is no roadmap that I'm aware of for this.
There's definitely been a lot of interest in this feature over the years from the community at large. Stanford had an approach very early on where they had pseudo-users that were dynamically generated and had enough information for XBlock mechanisms to work. This worked, though it led to a lot of junk in the auth_user table that needed to be cleaned up at regular intervals. It's possible that we could fit a layer of indirection here so that a User for the purposes of XBlocks doesn't map directly to a User in Django, and prevent auth_user from becoming a mess.
Session-backed storage for XBlocks has been the solution talked about more internally at edX in the past. We already do something like this for Studio, so that you can interact with problems that use session state while previewing the problem. This was actually supported in the LMS very early on in the platform (5+ years ago, back when it was still mitx), but I think it was broken sometime during the scramble of the first year. One issue here is what a transfer of data would look like, since it's basically a KV store and memcached doesn't have any indexes or nicer ways to organize the data by default when we want to do the "let's get all the user's answers and move them over". It's possible that we could use something like Redis here, or place restrictions on the data that we have to transfer over to simplify things.
Then there are the more policy-oriented questions around the control that course teams have over anonymous access to their content -- opt in vs. out, blocking sensitive parts because of licensing agreements or to help preserve exam integrity, etc. I don't know of any central place where these requirements are gathered.
Take care.
Dave