Assessing Omitted Variable Bias when the Controls are Endogenous