[diffusion] perf: add QKV fusion optimization for Flux models (#14505)

Co-authored-by: Mick <mickjagger19@icloud.com>

python/sglang/multimodal_gen/runtime/loader/component_loader.py

@@ -691,6 +691,11 @@ class TransformerLoader(ComponentLoader):
         ), "Model dtype does not match default dtype"
 
         model = model.eval()
+
+        if hasattr(model, "fuse_qkv_projections"):
+            logger.info("Fusing QKV projections for better performance")
+            model.fuse_qkv_projections()
+
         return model