Feature request: Update paddle OCR models to V4

[grinco]

Hi, looks like paddle OCR version 4 brings in increased detection accuracy, and it would be great if you upgraded the current models to use it.

I tried using it as a drop in replacement for en_PP-OCRv3_det_infer_rk3588_unquantized.rknn and en_PP-OCRv3_rec_infer_rk3588_unquantized.rknn but failed:

12:50:56:ALPR_adapter.py: E RKNN: [10:50:56.027] 6, 1
12:50:56:ALPR_adapter.py: E RKNN: [10:50:56.028] Invalid RKNN model version 6
12:50:56:ALPR_adapter.py: E RKNN: [10:50:56.028] rknn_init, load model failed!
12:50:56:ALPR_adapter.py: [ERROR] fastdeploy/runtime/backends/rknpu2/rknpu2_backend.cc(123)::Init	Load model failed
12:50:56:ALPR_adapter.py: [ERROR] fastdeploy/runtime/runtime.cc(328)::CreateRKNPU2Backend	Failed to initialize RKNPU2 backend.```

[grinco]

The error above went away after I replaced librknnrt.so - but there still are issues...