Vectorize rgroup get_weight_constraint

[dgcnz]

I'm too lazy to switch branches right now but this is the solution for (num_filter_banks, groups):

    def get_weight_constraint(w: torch.Tensor):
        w = w.unsqueeze(2)
        return torch.triu(w - w.transpose(1, 2)).abs().sum()

[dgcnz]

@Nesta-gitU

[dgcnz]

done