Ineligibles and Eligible Non-Participants as a Double Comparison Group in Regression-Discontinuity Designs