Fix decoder prefill

Fix model name...

Delete onnx/onnxrun.py

Delete onnx/rknnrun.py

Add a better inference script.

Delete onnx/vision_encoder_part1.onnx

Delete onnx/vision_encoder_part2.rknn

Update models and scripts with toolchain version 2.3.2

Update preconverted models

Delete onnx/vision_encoder.rknn

Split part of vision encoder to CPU and optimize Transpose ops. (Reupload to correct path)

Split part of vision encoder to CPU and optimize Transpose ops. (Reupload to correct path)

Fix RKNN data layout

Upload 21 files