Move expensive part of kinematics after cut evaluation

[kilian]

Currently, all kinematics and phase space calculations are completed before cuts are evaluated. If the phs point passes cuts, the matrix element is evaluated.

It turns out that the most expensive part of the phs calculation is the evaluation of the integration parameters x_i for all channels other than the current channel. This evaluation uses the momenta that are computed in the current channel. Therefore, this part can be moved after cut evaluation, since the latter needs just the momenta.

I expect that the first few iterations, where the cuts are missed most of the time, get a significant speedup, and OpenMP scales better, if this is implemented. First tests confirm this.

[kilian]

The modifications are implemented in r3235. The old version can be reactivated by setting ?old_phs_version=true. This allows comparisons and checks. If we agree that everything works, I will remove the old version.

[Juergen Reuter]

SS and JRR have not up to now discovered any deteriorations of the integration behavior compared to the previous setup.

[Juergen Reuter]

Just for testing and triggering an email....

[kilian]

No problems so far, therefore I remove the flag ?old_phs_version, so the new version is always active (r3267).