Wrong results with steaming=True when concat dataframes.

[jhirsch-mhp]

Checks

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of Polars.

Reproducible example

import polars as pl

print(f"polars version: {pl.__version__}")

# define simple Dataframe
d = pl.DataFrame({"a": [0,1,2,3]}, schema={"a": pl.Float64})
print(d)

# Make lazy
d = d.lazy()

# split, process and combine again
combined = pl.concat(
    [d.filter(pl.col("a") < 2).with_columns(pl.mean("a")), 
     d.filter(pl.col("a") >= 2)],
    how="vertical"
)

# Show Plans
print("\nPlan:\n", combined.explain())

# See different results for Streaming
print(f"\nStreaming=False: {combined.collect(streaming=False)}")
print(f"Streaming=True: {combined.collect(streaming=True)}")        # <--- This result is wrong

Log output

polars version: 1.20.0
shape: (4, 1)
┌─────┐
│ a   │
│ --- │
│ f64 │
╞═════╡
│ 0.0 │
│ 1.0 │
│ 2.0 │
│ 3.0 │
└─────┘

Plan:
 UNION
  PLAN 0:
     WITH_COLUMNS:
     [col("a").mean()] 
      FILTER [(col("a")) < (2.0)] FROM
        DF ["a"]; PROJECT */1 COLUMNS
  PLAN 1:
    FILTER [(col("a")) >= (2.0)] FROM
      DF ["a"]; PROJECT */1 COLUMNS
END UNION

Streaming=False: shape: (4, 1)
┌─────┐
│ a   │
│ --- │
│ f64 │
╞═════╡
│ 0.5 │
│ 0.5 │
│ 2.0 │
│ 3.0 │
└─────┘
Streaming=True: shape: (2, 1)
┌─────┐
│ a   │
│ --- │
│ f64 │
╞═════╡
│ 2.0 │
│ 3.0 │
└─────┘

Issue description

When using a lazy split -> process -> concat expression, the streaming=True yields a wrong result. It completely drops the part where an expression was done. This is not limited to the mean() function but is also true for any other expression.

PS; I also opened #20694, however I realized that this problem is much more generic then described there and applies to every split -> process -> concat expression. It is therefore much more general and maybe worth a fix. I tagged the other issue as duplicate of this one.

Expected behavior

Same behavior for streaming=True as it is for streaming=False.

Installed versions

--------Version info---------
Polars:              1.20.0
Index type:          UInt32
Platform:            Windows-10-10.0.19045-SP0
Python:              3.10.10 (tags/v3.10.10:aad5f6a, Feb  7 2023, 17:20:36) [MSC v.1929 64 bit (AMD64)]
LTS CPU:             False

----Optional dependencies----
Azure CLI            <not installed>
adbc_driver_manager  <not installed>
altair               <not installed>
azure.identity       <not installed>
boto3                1.35.7
cloudpickle          3.0.0
connectorx           <not installed>
deltalake            <not installed>
fastexcel            0.11.6
fsspec               2023.12.2
gevent               <not installed>
google.auth          2.34.0
great_tables         <not installed>
matplotlib           3.9.2
nest_asyncio         1.6.0
numpy                1.24.4
openpyxl             3.1.5
pandas               2.2.2
pyarrow              13.0.0
pydantic             <not installed>
pyiceberg            <not installed>
sqlalchemy           2.0.32
torch                2.4.0+cpu
xlsx2csv             0.8.3
xlsxwriter           3.2.0```

</details>

[orlp]

My reply is the same as in the previous issue:

This works correctly in the new streaming engine, I'm not sure if it's worth the time to fix this in the old streaming engine if it will be replaced soon.