fix: unpickling-error-pytorch-2.6

[NebulaMao]

issue: #71

This pull request improves the robustness of the YOLO model loading process in modules/subcore.py by adding compatibility for different PyTorch versions. The main change is the introduction of a fallback mechanism to ensure that the model loads correctly, regardless of whether the newer safe_globals feature is available.

Model loading compatibility improvements:

• Updated the load_yolo function to first attempt loading the model using PyTorch 2.6+'s safe_globals context, and if that fails (due to missing features or incompatible types), it falls back to the previous loading method. If both attempts fail, the error is logged and re-raised for better debugging.