Add the anatomix ViT encoder, trace both anatomix networks in evaluation mode

Move input normalization to models and make layer selection part of forward()

Initial upload of TorchScript models for IMPACT