v1.4.0

• Add torch support!
• Add .cpu(), .cuda(), .to() methods to move the dataclass from
  devices when using torch.
• Breaking: @dataclass(frozen=True) is now automatically applied