many-examples: remove kaggle dependency

[alexcg1]

As discussed in various meetings with @lusloher , @aga11313 , @FionnD

Kaggle is a lot of hoops for a user to jump through just to get an example working: install, set up key, run data getter script.

It's also work for us: We have to ensure datasets haven't moved or changed a lot, and we sometimes have to perform extra steps to process them.

These datasets are generally under creative commons licenses or similar. There's no reason why we can't:

• Download a subset for example purposes (this keeps things light)
• Process that subset ourselves (saves users time and effort)
• Store it either in data/ (for light stuff like text which can go directly in repo) or use get_data.sh to download from somewhere we control (for larger stuff like images)

Affected examples

• [ ] wikipedia-sentences
• [ ] multires-lyrics-search
• [ ] cross-modal-search
• [ ] query-while-indexing

[FionnD]

Thanks for creating the issue Alex!

Just to clarify to any engineer. ⚠️This issue should not be worked until https://github.com/jina-ai/examples/issues/447 and https://github.com/jina-ai/examples/issues/512 are completed. ⚠️

[nan-wang]

audio-search has no longer dependency on kaggle

[jakobkruse1]

Where could we store the example data? Do we have "somewhere we control" to download from?

[tadejsv]

I propose to use, when possible, huggingface datasets. They are extremely easy to use, and very performant too.