On the Adaptivity Gap in Two-Stage Robust Linear Optimization under Uncertain Constraints