[Tracker] Training and inference tutorials

Create a series of training and inference tutorials exploring Kornia API with up-to-date real cases open source:

- using hugging faces datasets library and datasets available on hugging faces hub
- using torchmetrics and/or evaluation for model evaluation
- using kornia x, lighting trainer, or other api with accelerate integration for training
- using kornia for data augmentation, exploring augmentation sequential api
- using timm for encoders

Exploring image classification, semantic segmentation, and object detection. Showing how to export and/or compile the pipelines for better performance

My goal here, aside from providing tutorials for the community, is to explore and identify inconsistencies between other libraries and Kornia. We can use these tutorials to explore improvements for kornia to be easier to use along other ML libraries.

Tracker list:

- training
    - [ ] image classification
    - [ ] semantic segmentation
    - [ ] object detection

- inference
    - [ ] image classification
    - [ ] semantic segmentation
    - [ ] object detection

Each tutorial will need to be skipped run on CI or be able to just use some samples. Open to discussion to what other libraries of the ecosystem we should have examples on tutorials

Start from the training and inference tracker and decide which single tutorial scope can be taken first: image classification, semantic segmentation, or object detection. Review the listed integrations with the Hugging Face datasets library, torchmetrics or evaluation, Kornia APIs, accelerate, and timm. Done means a tutorial uses real or sample data, covers the selected pipeline, and is skipped or runnable in CI.

Written by the indexing model from the issue text.