Add binned shower

This adds a binned shower class that is able to use voxelized GEANT4 samples in the form of HDF5 datasets to improve the accuracy FastCalorimeterSimulation. The energies of the individual voxels are mapped to cells by sampling several hits for each voxel and using the standard FCS HitToCell matching.

A phi-modulation class is used to correct for the dependence of the energy response on the hit position, relative to its cell. This class works with the usual FCS Simulation chain and is usable by other FCS modules as well.

Furthermore, a BinnedShowerONNX class is added as well, that removes the necessity of loading actual GEANT4 samples by using generative machine learning models instead. It uses the same simulation skeleton as the original BinnedShower. The ONNX models are loaded using the already existing ONNXHandler class.

Some other FCS modifications were needed to ensure thread safety.