GammaLib

While reworking the model module interfaces (see #773), I recognized that GModelSpatialDiffuseCube is not a spatial model component but a spatio-spectral model component. This could not be used in the actual scheme, as all interfaces are defined for factorized sky models. See the discussion at https://cta-redmine.irap.omp.eu/issues/773#note-10.

The interfaces should be re-considered so that also non-factorized sky models are supported. It would be nice if GModelSky could achieve this, without introducing derived classes for different sky models. So although the sky model should be composed of components, the interfaces should not reflect the factorization.

The problem can be resolved by passing all true attributes (GSkyDir, GEnergy, GTime) as arguments to the three model component eval methods. The model is still factorizes, but now in a generic way. The component separation is now per unit.

Argument passing should be done using the GPhoton class, which allows transparent evolution of the interface (e.g. when polarization needs to be added).

The GModelSky::value and GModelSky::gradient should also take GPhoton as argument, the same is true for the GModelSpatial::eval, GModelSpatial::eval_gradients, GModelSpectral::eval, GModelSpectral::eval_gradients, GModelTemporal::eval and GModelTemporal::eval_gradients methods.

The diffuse map cube has now been implemented (see #576), hence this action can be considered as solved.