fix: detect Z-Image LoRAs with transformer.layers prefix (#8986)

OneTrainer exports Z-Image LoRAs with 'transformer.layers.' key prefix
instead of 'diffusion_model.layers.'. Add this prefix (and the
PEFT-wrapped 'base_model.model.transformer.layers.' variant) to the
Z-Image LoRA probe so these models are correctly identified and loaded.

Author: Pfannkuchensack

invokeai/backend/model_manager/configs/lora.py

@@ -711,6 +711,8 @@ def _validate_looks_like_lora(cls, mod: ModelOnDisk) -> None:
             state_dict,
             {
                 "diffusion_model.layers.",  # Z-Image S3-DiT layer pattern
+                "transformer.layers.",  # OneTrainer/diffusers prefix variant
+                "base_model.model.transformer.layers.",  # PEFT-wrapped variant
             },
         )
 
@@ -747,6 +749,8 @@ def _get_base_or_raise(cls, mod: ModelOnDisk) -> BaseModelType:
             state_dict,
             {
                 "diffusion_model.layers.",  # Z-Image S3-DiT layer pattern
+                "transformer.layers.",  # OneTrainer/diffusers prefix variant
+                "base_model.model.transformer.layers.",  # PEFT-wrapped variant
             },
         )