Sorry @rufuspollock apparently notifications were disabled. Love this topic though. Where do you imagine the CSV would be persisted? GitHub? If so, I had a similar idea about essentially decoupling the "editing" functionality of JKAN into its own project -- kind of like prose.io, but for editing data. Each directory of data files could have its own _schema.yml file or something, and this service would let you browse the github repo and render a form for each data file based on that schema.
I rambled about that a bit on this thread a couple years ago: prose/prose#934 In retrospect I'm not sure it needs to be in prose.io
Of course, that still leaves open the question of the schema definition language. I really like table schema (we use it for all our ETL at the city of philadelphia), but I don't know whether it allows for new, HTML/form-related properties to be added?