PowderN 3.0 does not partition stats correctly

There are problems with the way PowderN distributes statistics between the 3 branches: coh, inc, and transmit/tunnel. In lines 654-657 McStas (683-686 McXtrace), the renormalization of p_inc and p_transmit uses the already renormalized p_interact. This leads to weightings which don't add up to 1, which is clearly wrong.

Furthermore the algorithm suggests a different concept than before - with this idea one should be able to say for instance 1:2:8 for the three relative branches but the docs say that p_interact =1-p_inc-p_transmit which is not neccesarily the case.

The new algorithm should be corrected and the docs should be updated.