Falsification of Internal and External Validity in Observational Studies via Conditional Moment Restrictions