I think we can treat the DB as an internal component. We can ensure data integrity through testing.
I really like the idea of validation with Zod because it's convenient and effective. However, I suggest we continue validating the endpoints (client-to-server communication) and not the server-to-database communication, as it's an additional step in the process of creating a simple database query.
Another point against validating server <-> database communication is the runtime overhead (we want to be blazingly fast).

I'll reconsider my decision when we release the project and hit over 1k stars on GitHub. :)

In general this is not a bad idea. However, the main points of validation should be on the client and when data is passed to our internal API (if there is any). In many cases, the data model for passing data to service layer in order to be persisted differs from the actual DB schema. Therefore, I do not really see the big advantage of this generation.
I really see a big advantage when when creating service layer and client side schemas and infer models from them.