Yes you can. Ergonomically that'll end up being harder sometimes. Sometimes it might be easier too! E.g. mostly loose data but you need a user_id field.
Not sure about performance. I imagine if that user_id field is a column you might get better performance. Not least because by choosing columns in a query you reduce the amount of data to be processed.
Another benefit of doing it the boring way is tooling. Orms, schema migrations and so on. I work in this space somewhat and think alot about the DX of both key value and relational stores!
Think of me as a "Just use postgres" and "Also just use dynamo" at the same time :)