change representation forward to allow fast forword if grad is not required
 add api to get latent variables of MLP
 some script to evaluate uncertainty

 dropout
 nn ensemble
 nndnn

fix some bug in standrize of energy
add test for kalman filter train and standrize
fix bug when prior is None and device is gpu

add standardize of energy
add AtomRefPrior
fix bug in BehlerG3 in gpu

add goedecker net

change loss_fn
add element for every SF