Adding dataset downloading to kagglehub

[jeward414]

This is part 1 of adding dataset support to kagglehub. This adds the functionality of downloading.

kagglehub.dataset_download("jeward/testDataset")
Optional file path -> kagglehub.dataset_download("jeward/testDataset", "foo.txt")

b/313706281

[rosbo]

In a follow up PR, can you add support in the Kaggle notebook resolver: https://github.com/Kaggle/kagglehub/blob/main/src/kagglehub/kaggle_cache_resolver.py

You will need to make a backend change first to add support for attaching datasets dynamically.

1. Add a DatasetReference to the oneof datasource_reference field here: https://github.com/Kaggle/kaggleazure/blob/2308f0d1d087c4b19946024aed8a6306f1bbf4ac/Kaggle.Sdk/kernels/kernels_service.proto#L2998
2. Implement the logic to dynamically attach a dataset: https://github.com/Kaggle/kaggleazure/blob/457303918d98b8a0d0931ce2ffbc99896253df5c/Kaggle.Services.Kernels/Handlers/AttachDatasourceUsingJwtHandler.cs#L73

The easiest way to test it in the kagglehub integration in a Kaggle notebook environment is through these steps:

1. Push your changes to GitHub on a branch.
2. Open a Kaggle notebook on kaggle.com
3. Run:

!pip install --force-reinstall --no-deps git+https://github.com/Kaggle/kagglehub.git@NAME_OF_YOUR_BRANCH

import kagglehub
kagglehub.dataset_download("jeward/testDataset")

[rosbo]

Do we want to add versions/ prefix or not?

Pros of keeping it:

• Matches the dataset URL.

Cons of keeping it:

• Handle is different from models.

@jplotts any thoughts?

[rosbo]

LGTM. Great work. Just a few small suggestions.