tutorial notebook for train UNET with cordexbench data

use cordexbench SA data to train UNET using AI downscaling code

Resources
- [CORDEX ML bench](https://github.com/WCRP-CORDEX/ml-benchmark)

### Tasks
- [ ] load cordexbench data
- [ ] create pytorch dataset for cordexbench
- [ ] compare train and test datasets
- [ ] create train pipeline - normalise, train
- [ ] create evaluation pipeline, train and test, metrics, vis
- [ ] train on CPU and GPU

Start with the CORDEX ML bench resource and the repository's AI downscaling code, then determine where the tutorial notebook should live. Work through loading CORDEXBench South Africa data, dataset creation, train/test comparison, training and evaluation pipelines, and CPU/GPU execution. Done means the notebook covers the listed tasks with metrics and visualisations.

Written by the indexing model from the issue text.