refactor: split and simplify sample_k_diffusion samplers (#1377)

refactor(server): split server endpoint registration (#1376)

fix: disable extra T5 mask padding for Wan (#1375)

refactor: move VAE tiling parameters to SDGenerationParams (#1261)

sync: update ggml

refactor: migrate generation pipeline to sd::Tensor (#1373)

refactor: simplify f8_e5m2_to_f16 function a little bit (#1358)

refactor: simplify sample cache flow (#1350)

perf(z-image): switch to fused SwiGLU kernel (#1302)

style: remove redundant struct qualifiers for consistent C/C++ type u…

…sage (#1349)