adding local observables

[dcyang]

Hi,
I've been experimenting with general (non-Coulombic) fermions and I'm currently trying to add g(r) (radial distribution function) to my branch, but the train() -> fit_wf() call from the main loop seems to (from its naming) hint at the main devs' intention to always involve compute_local_energy().

So I'm working with a clone of fit_wf() in fit.py that is modified to avoid compute_local_energy() in the steps accumulating the g(r) histogram.
Would you consider this to be in line with the future development direction of DeepQMC? (Or is there a more "modular" way to add these observables?)