A Boundary-Respecting Confidence Interval Method for a General Effect Size Measure in Paired Experiments