Currently we rely on MongoDB connector's schema inference, which is based on sampling. Add the possibility to explicitly provide a schema.