Add Geant4 EM model wrappers for Celeritas interactions

[sethrj]

One way we could improve our V&V process (and start integrating into frameworks) would be to define a G4VEmModel classes that would individually replace those in Geant4 and call the Celeritas interactors. It would be straightforward to pass our particle parameters, define a wrapper function for the CLHEP RNG, and convert the "interaction" result to a "particle change".

This could be especially useful for the muon capture/fusion models, where it might be a long time before we can start validating an entire muon beam simulation. (@stognini )

One difficulty is that we might need to subclass the corresponding Geant4 model so that we can rely on its cross section generation (and our import process), and then switch out its "SampleSecondaries" functionality at runtime.

This would sort of be the reverse of the "pull G4HepEm into Celeritas" approach: but since our samplers are more general/flexible it might be easier to do.