SupSiam: Non-contrastive Auxiliary Loss for Learning from Molecular Conformers