[WIP] Local energy conservation loss

[oviazlo]

The first prototype of the function to calculate local energy conservation loss. To discuss.

[jkiesele]

Thanks. In the end we'd need to put it in another file, together with other losses maybe; as I am trying to clean up a bit.
But this can be done even after the final PR is merged, since now it's quite useful to have it separate :)

I just had a quick look, but I am getting more and more the feeling that we should save the deposited true energy directly in the dataset - would that help you?

[oviazlo]

Hi @jkiesele. I rewrote the function. I hope this time I got everything correct. I was inspecting the output of the function and was comparing it with ediff calculated in LossLayers.py:
ediff = (t_energy - pred_energy*dep_energies)/tf.sqrt(t_energy+1e-3)
Both matrices were of a similar order.