Release 0.2.0

This release enables a refactor that now actually uses structured outputs.
We also now are using gpt-4o by default without relying on free models as providers might train on that data.