XFormerXTrans2 and XFormerDenoise do not work with ONNX due to float16 vs float mismatch

[patrakov]

[aep@dell-laptop 105NIKON]$ ~/AI/rawforge-venv/bin/rawforgeonnx --cfa --onnx XFormerDenoise DSCN0152.NRW DSCN0152-xformer-denoise.dng
0it [00:00, ?it/s]
Traceback (most recent call last):
  File "/home/aep/AI/rawforge-venv/bin/rawforgeonnx", line 7, in <module>
    sys.exit(main())
             ~~~~^^
  File "/home/aep/AI/rawforge-venv/lib/python3.14/site-packages/RawForge/main.py", line 152, in main
    run_pipeline(
    ~~~~~~~~~~~~^
        model_names=args.model,
        ^^^^^^^^^^^^^^^^^^^^^^^
    ...<14 lines>...
        disable_tqdm=args.disable_tqdm
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/home/aep/AI/rawforge-venv/lib/python3.14/site-packages/RawForge/main.py", line 68, in run_pipeline
    _, output_img = worker._tile_process(output_img, handler.model_params, progress_callback=progress_callback)
                    ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/aep/AI/rawforge-venv/lib/python3.14/site-packages/RawForge/application/InferenceWorker.py", line 78, in _tile_process
    output = self.model.run(["output"], filtered_inputs)
  File "/home/aep/AI/rawforge-venv/lib/python3.14/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 322, in run
    return self._sess.run(output_names, input_feed, run_options)
           ~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Unexpected input data type. Actual: (tensor(float16)) , expected: (tensor(float))

And the same for XFormerXTrans2. The Torch-based path is too slow for me to test.