and there's no distinction between a missing column and one that didn't exist, how does hashing work?
Same way as before. I don't see that as a practical stumbling block, but maybe you have a specific use-case in mind that would muck things up?
Informal categorisation and structuring has its place, but that's an entirely different beast to a relational database.
Indeed with regard to informal structuring: something easy to get going is often useful for prototyping. One can then lock down this tool incrementally as things settle (or migrate to a static RDBMS).
I've been in rather long debates about the definition of "relational database", and found no clear-cut "failure" to match. Language is subject to interpretation.
Anyhow, the idea is to produce a useful tool. It's formal category or definition is secondary to being useful.