Parallelize the sampling functions using OpenMP

[kilian]

The sampling function is responsible for the largest part of the Whizard computing time. It should be run in parallel mode for efficiency.

Currently, the helicity loop can be run in parallel via OpenMP (cf. #380), which allows scaling up to about 4 cores. However, about 20-40% of the CPU time are spent in other parts of the sampling function. These should also be OpenMP-ized:

1. The procedure phs_forest_evaluate_other_channels: can be evaluated in parallel for O(10-1000) channels.
2. The procedure combine_particles: rather trivial, but takes time. May be optimized without parallelization
3. The procedure vamp_probability. Here, the called subroutine divisions / probability_s by itself can use up 20% of the total running time of the program.

[Juergen Reuter]

Would make sense to handle #234 before that one here...

[kilian]

Done in r3301.

All credits go to Matthias Trudewind.