Agree, and no users from Sub-Saharan Africa either - although both regions have an extremely vibrant and ambitious Wikimedia community.
I notice that I'm already spending (much) more time in general than the hours allocated / budgeted in the current Wikimedia grant. It's very labor intensive work, also because I need to onboard myself into a new platform (WikiLearn based on Open edX). That said, budgets are limited, and the number of people and their capacity to do this too. I'd personally prefer to spend my time doing actual work with (tools like) OpenRefine rather than training. What's the best bang for the (few) buck so that more people can actually work with data, and so that we need less time and money watching and creating videos?
There is indeed already "so much stuff" out there. It's indeed a challenge, and time consuming, to watch, read and pick the best for the job. One example: there are many online explainers for the cell.cross function but the single one that I find enlightening is this one due to its color coding and very visible explanations in the screenshots. However, it rarely drifts to the top in my search engine results and I always am looking for it again! A big issue is keeping things up to date age as OpenRefine changes. Some documentation that I created last year already needs/needed updates. I'm hesitant about the video creation because a few changes in the application may already render a video useless (after it was time consuming to produce).
Some type of in-app guide would be absolutely amazing, a real time saver. It would basically take away a lot of the need for the repetitive "how does OpenRefine work, what does everything mean, are all the buttons and menus" dance that each trainer now does. It would also make users feel much more confident that they're doing the right thing when trying the application themselves.
I found it very interesting that this drifted to the top. But yes, every group of trainees asks for this in some form. Basically, folks have some other dataset, either big or small, that they want to "fuzzily" compare with the main one they work with. I'd say it's what csv-reconcile and reconcile-csv try to do. Now that one (csv-reconcile) has a complex installation process (I'm not dumb, but I don't understand how to do it, and so I also don't actively recommend it to people I train and work with.) Compared to that, installing the Commons extension is a breeze 
I'm surprised too. I don't have strong opinions, I think this really needs very in-depth user research.
In the various scenarios in which I use OpenRefine, it would only be useful if it comes with functionalities very similar to Google Docs: the ability to communicate with collaborators, tag them, explain what's been done at a certain point, tell them where to continue, what to repeat, etc.