feat: Add cs.remaining() selector

[gab23r]

Towards #12067

Add cs.remaining() selector that selects columns whose output names are not produced by other expressions in the same select() or with_columns() call.

Example:

import polars as pl
import polars.selectors as cs

df = pl.DataFrame({ "b": [4, 5, 6], "c": [7, 8, 9], "a": [1, 2, 3], "d": [10, 11, 12]})

# Reorder columns
df.select("a", cs.remaining(), "d")
# shape: (3, 4)
# ┌─────┬─────┬─────┬─────┐
# │ a   ┆ b   ┆ c   ┆ d   │
# │ --- ┆ --- ┆ --- ┆ --- │
# │ i64 ┆ i64 ┆ i64 ┆ i64 │
# ╞═════╪═════╪═════╪═════╡
# │ 1   ┆ 4   ┆ 7   ┆ 10  │
# │ 2   ┆ 5   ┆ 8   ┆ 11  │
# │ 3   ┆ 6   ┆ 9   ┆ 12  │
# └─────┴─────┴─────┴─────┘

The implementation uses a two-pass approach in rewrite_projections:

1. Expand all non-Remaining expressions and collect their output column names
2. Transform Selector::Remaining → Selector::Wildcard - Selector::ByName(output_names)

I used AI to explore different implementations and I confirm that I have reviewed all changes myself, and I believe they are relevant and correct.

[gab23r]

@alexander-beedie, curious about your thoughts as you already think about adding this selectors !

[coastalwhite]

I would personally be quite uncomfortable with such a two-phase approach specifically for this selector. Like, how would this work at other places that are not with_columns or select? The two-phase approach was how selectors used to be implemented and caused a large amount of messiness, unpredictability, and just general bugginess.

For group_by(...).agg(...), cs.all() already doesn't include all columns. Here it is implemented through the ignored_selector_columns. Maybe such a mechanism can be altered or reused somehow?

[gab23r]

I battle test it more and you were right, I had some bugs with selectors athimetics, I fixed them but now the logic starts to be quite messy... I will try to revisit using the already existing ignored_selector_columns

[gab23r]

I now use the already existing ignored_selector_columns which is much better than doing this transformation: Remaining → Wildcard - ByName(output_columns).

[gab23r]

@coastalwhite, maybe you wanted the logic to happen before rewrite_projections like in the group_by case.
In this cs.remaining case, I don't see how to do that without modify a lot a logic a bit of everywhere. And I still think that applying the logic here make more sense.

I have battle test the implemention with other context than select or with_columns and I don't see issue and I can't imagine any potential futures bugs. But maybe you have more precise ideas where bug can show up here ?

I have added the unstable tag to cs.remaining.