Implement T-guess for fixedT problem

[timmercy]

When solving a fixedT problem you have to select a certain guess for the total motion time. If you make an unrealistic guess the problem won't solve well. E.g. if you select your T too low the trajectory can be suboptimal because it doesn't take into account obstacles which are situated beyond the time horizon.

Possible implementation is to generate a guess via a freeT problem and use the resulting T as initial guess for the fixedT problem.