Improve performance of dataset Logger

[AyushExel]

This PR discards unnecessary operations performed on labels before logging which should speed up things for large datsets.

@glenn-jocher there's still a bug in this which I couldn't explain properly over mail, let me try again in the meeting today.
Before the images are logged, the dataset directory is registered as artifacts, so we need to log the images from the registered path to make sure they don't get duplicated.

The bug that I'm seeing is that this logging code works perfectly when the dataset consists of all square images(case1) but the boxs are misplaced in cases where images are raw/unaugmented(case2)
See Example output -> case 1-> augmented dataset, case 2-> raw dataset
[I have marked the exact logging code L245-L255 in wandb_utils.py]
To reproduce:
python utils/wandb_logging/log_dataset.py this logs raw coco128 dataset as W&B Table with slightly misplaced bboxes.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Improvement in Weights & Biases (wandb) dataset artifact logging in the YOLOv5 repository.

📊 Key Changes

• Updated LoadImagesAndLabels calls to use rect=True and batch_size=1 when creating dataset artifacts for training and validation sets.
• Simplified the bounding box data creation by using middle point (x, y) and dimensions (width, height) instead of previous corner points (minX, minY, maxX, maxY).
• Removed unnecessary attributes such as scores and domain.

🎯 Purpose & Impact

• These changes are intended to make the dataset artifact logging process for wandb more efficient and the bounding box data more intuitive.
• Enhancing the dataset artifact creation can potentially lead to easier integration with other tools and clearer visualization in the wandb interface.
• Users of YOLOv5 leveraging wandb will find it simpler to review and understand their object detection dataset metrics due to these modifications. 🚀