Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries