Finite Population Regression Adjustment and Non-asymptotic Guarantees for Treatment Effect Estimation