Whenever I write a backend, it's a thin layer on top of the RDBMS like you said. I don't know if a lib or framework could help with this. It's more about designing the schema well, avoiding excessive tooling (ORMs, query builders, etc), not trying to abstract away the DB, and writing ample integration tests.

If you get that stuff out of the way, you can focus on the real problems like design, xact isolation, and performance, for which there's tons of conflicting advice rather than an agreed-upon approach. And then there's sharding. It's hard enough already.

Personally I haven't found the need for type safety in code, or even the code-SQL boundary. The DB tables have types, as does my OpenAPI or Protobuf or whatever API spec. That's basically everything already. If something slips past my tests, it's because the tests are bad, and stronger typing wouldn't have helped.