Sensitivity analysis for the average treatment effect in paired observational studies