Infer candidate data schemas from representative records while distinguishing observed fields from inferred types or optionality.
