enh: `Expr.filter` for spark-like backends

[MarcoGorelli]

For LazyFrames, we can support Expr.filter so long as it's followed by an aggregation. This should already be validated at the Narwhals level, I don't think anything needs changing there

For spark-like, Column.filter doesn't exist, but we can use F.expr to accomplish the same:

from sqlframe.duckdb import DuckDBSession
import sqlframe.duckdb.functions as F

df = DuckDBSession().createDataFrame(pd.DataFrame({'a': [1,1,2], 'b': [4,5,6]}))
df = df.select(
    F.expr('sum(b) filter (where a==1)').alias('c'),
    F.expr('sum(b) filter (where a!=1)').alias('d'),
)
df.show()

+---+---+
| c | d |
+---+---+
| 9 | 6 |
+---+---+

If anyone fancies implementing Expr.filter for _spark_like using the above, then my hope is that it will "just work", similar to how it currently works for Dask

[MarcoGorelli]

🤔 implementation-wise this might not be so simple though, as it involves parsing strings...uurgh

[MarcoGorelli]

In Ibis we can do

df.select(c=_.b.sum(where=_.a == 1), d=_.b.sum(where=_.a != 1))

which is the other way round compared with how we're looking to do it here, so that also won't be too simple

marking as 'low-prio', but perhaps we can get to it eventually

[lucas-nelson-uiuc]

I tried implementing something a while ago too but it involved collapsing-expanding columns as array objects. Don't mind looking into this after closing out a couple other issues