feat: improved model_exporter

[mgumowsk]

What does this PR do?

• Fixed EfficientNet-V2-L preset: corrected mean/scale values from ImageNet standard (123.675 116.28 103.53 / 58.395 57.12 57.375) to 127.5 127.5 127.5 / 127.5 127.5 127.5
• Fixed reverse_input_channels being set to false for one model preset — corrected to true
• Fixed Mask R-CNN label mapping: replaced the naive labels shift with a proper COCO 91-class → COCO 80-class lookup table (COCO91_TO_COCO80) so category IDs that COCO dropped are handled correctly
• Fixed Mask R-CNN preset: lowered confidence_threshold from 0.5 to 0.05 and changed labels from COCO_V1 to COCO_80
• Added ExportParams frozen dataclass to pytorch.py to bundle export/quantization parameters instead of scattering individual config.get(...) calls
• Extracted repeated boilerplate from all four converters (getitune, timm, torchvision, yolo) into shared BaseConverter helper methods

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?