Fix dylora create_modules error when training sdxl

[tamlog06]

Fix networks/dyrola.py so that it works with sdxl_train_network.py.

In the latest implementation, when defining self.text_encoder_loras inside the __init__ function of DyLoRANetwork in network/dylora.py, the create_modules function expects an input of type CLIPTextModel. However, the actual input is either a List[CLIPTextModel] or CLIPTextModel, leading to a bug when the input is a List[CLIPTextModel]. This has been corrected to align with the definition of self.text_encoder_loras in network/lora.py, allowing the text_encoder to accept inputs of type List[CLIPTextModel].

[kohya-ss]

I didn't test DyLoRA with SDXL. Thank you for this!