Add `Expr.scatter`

[wukan1986]

Description

Currently scatter is only available on Series. It would be more useful if it was also available on expressions.

[MarcoGorelli]

scatter takes two arguments: the indices, and the new values. I think you're missing the values from your example

going to cc @orlp on this one

[wukan1986]

may be

pl.full_like(pl.col('A'), fill_value=None).scatter(pl.col('A').arg_not_null(), pl.col('A').drop_nulls())

or

pl.scatter(pl.col('A').arg_not_null(), pl.col('A').drop_nulls())

import polars as pl

df = pl.DataFrame(
    {
        "A": [5, None, 3, 2, 1],
        "B": [5, 3, None, 2, 1],
    }
)

df = df.with_columns([pl.lit(None).alias("nulls")])
print(df)

df = df.with_columns([
    pl.col("nulls").scatter(pl.col('A').arg_not_null(), pl.col('A').drop_nulls()).alias("C"),
])

print(df)

[orlp]

I think allowing arbitrary expressions with scatter is useful, and shouldn't be too difficult to add. I'd like to keep the current signature though, so Expr.scatter(indices, values) where indices and values are equal length (or values has length one and is being broadcasted).

[orlp]

@wukan1986 I edited the original post to just focus on the feature request instead of the specific example.