'checkpoint.append_modal' incorrectly defines shift and scale values

`scaler = init_shift_scale(dst_config)` initialize scaler values from config, not model. (problematic when train_shift_scale = True)

For now, inserting below just before `return` bypass this problem.
```
for idx in range(orig_state_dict['rescale_atomic_energy.shift'].shape[0]): 
    new_state_dict['rescale_atomic_energy.shift'][idx,:] = orig_state_dict['rescale_atomic_energy.shift'][idx]
```