Header in the lossfunction logging incorrect when energy only

[Zachary Ulissi]

From what I can tell, the log headers aren't correct for energy-only training in the lossfunction logging (line 313-315, model/init.py). They don't have fields for the step/time and loss function, and don't match what gets outputted in check_convergence at line 685.

[Alireza Khorshidi]

Should be fixed here: 9c121db