Fix nested column flattening in group_by to use underscores

[nablabits]

This is an early attempt to fix #1329

I'm not very familiar with the code but with my knowledge and the help of some LLM I managed to put together this solution. I'm aware that this could be quite disruptive for users as it's changing the output of some well established function such as to_list (see the example in this test). People more knowledgeable of the use cases can judge this.

If we want to preserve the nested.level1.name structure, I feel that the options go through:

1. how the signals are extracted here
2. the to_pandas method and more precisely this get_headers_with_length folk (source) which is responsible of getting the columns that then become a multi-index.

Let me know what you think

(There are a couple of lint errors that we can address once a consensus on the approach is reached)

[shcheklein]

@nablabits PTAL https://github.com/datachain-ai/datachain/pull/1496/files . Could try to to run the use case on top of that PR?

[nablabits]

@shcheklein Yep, I have rebased the branch on top of #1496 and yes, it is working as expected, see: nablabits@1570d50

I will close this now, but feel free to reopen if something else comes up 🙂

Edit: wait, it's working because I didn't remove the flatten=True 🤦