Improve the CPU usage of the MM strips response simulation

With this MR the performance bottleneck in the MM strips (ATLASSIM-4923) response is resolved. The bottleneck was is in generating random numbers from the mixture of two gaussian distribtions used for the transverse diffusion. Using a different method to get these numbers the CPU time spent to generate these numbers went from about 300 s/evt to not showing up in the flame graph. Also the other random number generators have been moved away from TF1 were possible. For now TRandom3 is used, but this will be changed to use the random number service later. The total time spent in the MM digi is now down to 25 s/evt or about 30% of the pixel digi.