Enabling tiled decode for SDXL in OneTrainer

StableDiffusionXLLoRAModelLoader.py

# ... some lines from this file are left out ...
    def load(
            self,
            model_type: ModelType,
            model_names: ModelNames,
            weight_dtypes: ModelWeightDtypes,
    ) -> StableDiffusionXLModel | None:
        base_model_loader = StableDiffusionXLModelLoader()
        lora_model_loader = StableDiffusionXLLoRALoader()
        embedding_loader = StableDiffusionXLEmbeddingLoader()

        model = StableDiffusionXLModel(model_type=model_type)
        self._load_internal_data(model, model_names.lora)
        model.model_spec = self._load_default_model_spec(model_type)

        if model_names.base_model is not None:
            base_model_loader.load(model, model_type, model_names, weight_dtypes)
        lora_model_loader.load(model, model_names)
        embedding_loader.load_multiple(model, model_names)

        model.vae.enable_tiling() 

        return model