Refactor pytorch-related inference logic

[lyuzinmaxim]

For each refactoring ticket, we need to ensure that all affected scripts and modules continue to function correctly and produce identical results to the pre-refactoring version.
Run both the pre-refactored and refactored versions of the script, then compare their outputs (console logs, generated files, or results) to confirm there are no differences.

This also includes verifying that:

• Imports, file paths, and output directories are updated consistently across the project.
• Documentation accurately reflects the new structure and usage.

In this ticket, we need to refactor the pytorch-related inference logic. Please don't touch now onnx inference, pytorch->onnx conversion, and evaluation scripts. The proposed (feel free to suggest) dir structure looks like:

src/
    inference/
        base/
            classifier_inference_base.py
        pytorch/
            resnet_inference.py
            mobilenet_inference.py
        onnx/ # <- we don't touch it in this ticket
        utils/
            <FILE_NAME> # e.g., for _to_pil_image(...) image_conversion.py

• src/inference/base/classifier_inference_base.py contains only base class
• src/inference/pytorch/resnet_inference.py and mobilenet_inference.py contain only related classes
• utils functions needed for all those classes moved to src/inference/utils/, arranged into files based on logic
• process_images_with_oop(...) moved to src/inference/utils/ and renamed (we don't need to specify the OOP in the name)
• if needed, move the functions from __main__ block to a separate file in src/inference/utils/

DoD:

• refactored code working in the same way as previously, all docs updated (if needed)