SLA: Beyond Sparsity in Diffusion Transformers via Fine-Tunable Sparse–Linear Attention - View it on GitHub