feat(wandb): use artifacts to track datasets

[borisdayma]

Datasets can now be tracked by using WandbCallback(dasaset_path='my_data')

It uses W&B artifacts.

See example artifacts on camvid dataset.

The dataset is versioned and has aliases such as latest (custom ones can be used) and will identify any changes of dataset between runs. Different datasets can be used by using different names.

Here is a demo notebook.

TODO in subsequent PR:

• trained models should also be logged as artifacts for better tracking and to allow more flexible pipelines (instead of files uploaded from a run)

[review-notebook-app[bot]]

Check out this pull request on 

Review Jupyter notebook visual diffs & provide feedback on notebooks.

---

Powered by ReviewNB

[borisdayma]

Note: I'm using master branch of nbdev. Not sure if I should use the last released version instead for CI

[jph00]

Yes last released version is best - thanks!