Add Simulation module that returns the gradient for a specified variable on call

Short description of desired functionality:
Subclass of SimulationBase that overloads __call__ function with signature (dataset, variables_to_observe, **kwargs) and returns the gradients to the observed variables along with the simulated results.

New definitions:
class DifferentiableSimulation

Implementation Idea:
Overload __call__ function and use tf.GradientTape to record the gradients. To keep memory-footprint feasible manually accumulate gradients for the summation of the fourier encoding.

Potential conflicts:

Test concept: