Fix cuda error when sorting empty pl.concat result

[jberg5]

Description

Should close #21824

See test case for repro, but basically if you had

    df = pl.LazyFrame({"a": [1, 2, 3]})
    # Create two filters, both of which will give empty results
    q = pl.concat([df.filter(pl.col("a") == 0), df.filter(pl.col("a") == 4)]).sort("a")

then q.collect(engine="gpu") you'd get a cuda exception because we'd try to do an out of bounds access on an empty table.

edit: turns out #21690 fixed this issue. This PR now will only contribute a test case.

Checklist

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

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[mroeschke]

/ok to test 8737854

[jberg5]

Looks like #21867 moved the tests into legacy/ - not sure if this changes things, or if a massive rewrite is in the works? @mroeschke what do you think, is this fix still relevant?

[mroeschke]

Sorry for the delay @jberg5

On main, I believe we see the correct behavior now

In [1]: import polars as pl

In [2]: df = pl.LazyFrame({"a": [1, 2, 3]})

In [3]: q = pl.concat([df.filter(pl.col("a") == 0), df.filter(pl.col("a") == 4)]).sort("a").collect(engine="gpu")

In [4]: q
Out[4]: 
shape: (0, 1)
┌─────┐
│ a   │
│ --- │
│ i64 │
╞═════╡
└─────┘

So a unit test would still be appreciated (in python/cudf_polars/tests/expressions/test_sort.py would be fine)

[jberg5]

Thanks @mroeschke , just have the test case now.

[mroeschke]

/ok to test e032a1b

[mroeschke]

/merge

[mroeschke]

Thanks again @jberg5