Sqrt(dE/dx) instead of log(dE/dx)

[gwm17]

@sigmanotation observed that taking the sqrt(dE/dx) vs Brho shows much more significant improvement over log(dE/dx) vs Brho. There is even some theory background for why this works better (dE/dx prop to z^2/v^2, brho is prop to z/v -> linearize through sqrt i think is the logic, but I have to check).