All children of ParameterEstimationTask should be required

[luciansmith]

Every parameter estimation task should have a child algorithm, objective, and at least one adjustable parameter and fit experiment.

[luciansmith]

These are all now checked to make sure there's only one of the given children, but not that there are at least one of the given children. This is probably a more general 'add more validation' task, though.

[fbergmann]

Considering that SBML went in favor of allowing to encode partial models ... should we allow for partially setup tasks? But yes, more validation rules would probably be good.