Add fallback to HStack lowering in cudf-polars

[rjzamora]

Description

Closes #19150
(Although, we can probably do a bit more work to avoid fallback for some non-pointwise expressions)

We were not validating that the items in HStack.columns were all "pointwise" before lowering to partitionwise logic. This adds the possibility for single-partition fallback.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[TomAugspurger]

Thanks @rjzamora, looks great.

Do you think this could affect any other operations? Do you think we should do some assert in _lower_ir_pwise to validate that everything actually is piecewise (probably not, but maybe...)?

[Matt711]

I'm curious if the test is failing only with pytest-xdist turned on?

[Matt711]

Suggested change

	df = (
	pl.DataFrame({"dt": dates, "a": [3, 7, 5, 9, 2, 1]})
	.with_columns(pl.col("dt").str.strptime(pl.Datetime("ns")))
	.lazy()
	)
	df = (
	pl.LazyFrame({"dt": dates, "a": [3, 7, 5, 9, 2, 1]})
	.with_columns(pl.col("dt").str.strptime(pl.Datetime("ns")))
	)

[rjzamora]

Seems like engine='gpu' raises when I use this?

[wence-]

Yes, we don't support this strptime deducing the format I think. (cc @brandon-b-miller who implemented this stuff).

[rjzamora]

Ahh - Looks like we have a serialization/hashing error when the new test runs in distributed mode :/

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[wence-]

Tiny suggestions, but I think this looks good.

[rjzamora]

/merge