1) Yes. The wiki has been corrected. Thanks!
2) Lane changing for speed gain is performed after a vehicle acquires enough "motivation" by being slowed down. The lcSpeedGain parameter controls the motivation threshold value where lane changing occurs (threshold = c / lcSpeedGain). If the value is set very high than even small slowdowns by the lead vehicle will instantly trigger lane changing.
3) No. The speed gain motivation depends on the speed difference so it will keep growing as longs as the slow leader is being followed. However, you could set the speedGain parameter via TraCI depending on the situation to achieve the probability effect. Note, that opposites.guess is unrelated to this.
4) sigma > 0 models driver imperfection via random slow-downs (and by not accelerating as strongly as possible). This is important for jam-stability and introducing randomness into the simulation. It also has some impact on gaps but to calibrate gaps the tau parameter is suited much better. In different calibration studies different values of sigma showed up. 0.5 was selected as a compromise. There is no interaction with the other junction model parameters.
regards,
Jakob