based on contour integral, without apparent singularities.
Validated against old (Rémy/Céline) implementation: agrees,
except for strong deviations near q->0, where the new
implementation is smoother.
The analytical calculations are explained in App. C2 of the manual.
This resolves #1300.

Replace Monte Carlo MISER integrator with class with new interface (like IntegratorReal) and removed obsolete VEGAS integrator

Moved integrator of InterferenceFunction2DParaCrystal to private member (to not (de)allocate gsl workspaces all the time)

Simplified IntegratorReal (no void* argument needed anymore in function to integrate and encapsulation of the callback object)