chore: rewrite `agg_pandas` to use `NamedAgg`

[MarcoGorelli]

Since cuDF have introduced NamedAgg, we can probably refactor the agg_pandas function to use NamedAgg. It may simplify things a bit, and avoid going via duplicate column names

This would just be an internal refactor

[FBruzzesi]

That would be really nice!

My concern is that to end up with a dedicated path for modin, wouldn't we?
(I could not find anything documenting modin support for NamedAgg)

[MarcoGorelli]

looks like they have it

In [13]: import modin.pandas as mpd

In [14]: mpd.NamedAgg
Out[14]: pandas.core.groupby.generic.NamedAgg

[FBruzzesi]

Then... I am all in for it 😁🚀

[dangotbanned]

Hey @MarcoGorelli, I'd like to close this as Not Planned.

I battled with that API to try and force it to work, but in the end it just complicated things.
Hopefully you can get an idea from this commit how much extra work it was:

• refactor: Switch back from .agg(**named_aggs) to __getitem__

The main issues were documented here (#2680 (comment)), but the whole PR is full of chasing down edge cases all related to non-str columns 😅

[MarcoGorelli]

sure, let's do that then - thanks a tonne for having explored it! 🙏

[dangotbanned]

sure, let's do that then - thanks a tonne for having explored it! 🙏

No worries!

I doubt I would've ended up in the same place in #2680 without trying it, so thanks for the suggestion 🙂