Add scoring "observables" for comparison with ATLAS

[sethrj]

To compare with ATLAS FullSimLight, and as a diagnostic in general, we should add a diagnostic that produces a mean (per event) with an estimate of the error on that mean:

• energy deposit
• step length, binned by charged vs neutral
• number of steps , binned by charged vs neutral
• number of secondaries, binned by particle type

We should be able to produce these values:

• On CPU/GPU with an action
• Through celer-g4 through an adapter function when using pure Geant4

It should output to a JSON diagnostic.

[esseivaju]

Is this closed by #1664 ?

[sethrj]

No, I think we either want something with the particular features ATLAS requires (separation by particle type, etc) or rely on a stepping action feedback mechanism.