Creating dataframe from Python dict with array as value ignores/loses values when starting with None

[aostermeier-100]

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

    data = [
        {
            "measurement_location": "um03",
            "date": datetime.datetime(2025, 2, 27, tzinfo=TZ).date(),
            "consumption": [None] * 95 + [327.0],
        },
        {
            "measurement_location": "um04",
            "date": datetime.datetime(2025, 2, 28, tzinfo=TZ).date(),
            "consumption": [327.0] + [None] * 95,
        },
    ]

      return pl.DataFrame(data)

Log output

Issue description

When creating test data like above I have noticed that the elements in data are not converted into the dataframe as expected, for cases where they array does start with None. As soon as I have one leading value, which is not None the whole initialization is working properly.

um03 ┆ 2025-02-27 ┆ [null, null, … null]
um04 ┆ 2025-02-28 ┆ [327.0, null, … null]

Expected behavior

Independent of what data I am converting into a dataframe with polars I would expect that the values are not being dropped or getting lost in this conversion.

um03 ┆ 2025-02-27 ┆ [null, null, … 327.0]
um04 ┆ 2025-02-28 ┆ [327.0, null, … null]

Installed versions

Polars:              1.29.0
Index type:          UInt32
Platform:            Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.36
Python:              3.12.9 (main, Mar 18 2025, 03:28:39) [GCC 12.2.0]
LTS CPU:             False

[aostermeier-100]

we have conducted some more tests and it seems that 25 might be the magic number here - as you can see the on the screenshot the first array contains the value and the second one does not.

arr_broken = [{"key1": "value_1", "datetime": datetime.datetime.now(), "list": [None]24 + 72 [1.0]}, {"key1": "value_2", "datetime": datetime.datetime.now(), "list": [None]25 + 71[2.0]}

[cmdlineluser]

#22302 (comment) may serve as an alternative approach.

shape: (2, 3)
┌──────────────────────┬────────────┬───────────────────────┐
│ measurement_location ┆ date       ┆ consumption           │
│ ---                  ┆ ---        ┆ ---                   │
│ str                  ┆ str        ┆ list[f64]             │
╞══════════════════════╪════════════╪═══════════════════════╡
│ um03                 ┆ 2025-02-27 ┆ [null, null, … 327.0] │
│ um04                 ┆ 2025-02-28 ┆ [327.0, null, … null] │
└──────────────────────┴────────────┴───────────────────────┘