ToDo

When initializing a SmoothModelRun smr we give a startvector.
When we use smr to calculate an age density we usually give it a start distribution THAT IS NOT NORMALIZED
By integrating over all ages of this start distribution the masses in the pools could be inferred, so the information is duplicated.
One could possibly provide contradictory startvalues and startdistributions.
To avoid this the function providing startdistributions return a tuple that contains the startvalues belongin to the distribution,( which is silly) 

(Alternative) Proposels to solve this issue.
1.) If a startdistribution is available allow initialization of a modelRun with the start distribution only.(write an extra constructor)
    - disadvantage: Inference of the mass (if needed) by integration over the distribution is extremely expensive and potentially sligthly inaccurate

2.) Deal with normalized age distributions only, this does not even need an extension to the SmoothModelRun class 
    although would be possible to add a startdistribution argument to init.
    - advantage: The computation of solutions is extremely cheap. To make a distribution from a normalized density costs nothing.
     even the reverse is true (If we have solved for a real distribution we can divide by the pool contents which are available in the (cheap) trajectories)