The term is using a trick to avoid using the abs() function, which may otherwise trigger frequent zero-crossing when mdot is small. Using the sqrt() formulation and having a value that's always > 0, no zero-crossing would be triggered when solving the equations.
Later the momentum balance equation needs to compute abs(delta_vel_AI)*mdot_A/area. Here delta_vel_pos_AI is approximately abs(delta_vel_AI) when delta_vel_AI is not very small. If delta_vel_AI is small or close to 0.001*speed_of_sound, the mdot_A term would also be small, and the impact on the total momentum balance should also be neglible.
I'm curious what you found after modifying the block. Would you be able to share the .ssc file and a test model?