Downcasting temporal units silently overflows

[FHTMitchell]

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

In [13]: df = pl.DataFrame(
    ...:     [pl.Series("dt", [datetime.max, datetime.now()], dtype=pl.Datetime("us"))]
    ...: ).with_columns(td=pl.col("dt") - datetime.fromtimestamp(0))

In [14]: df
Out[14]: 
shape: (2, 2)
┌────────────────────────────┬───────────────────────────────┐
│ dt                         ┆ td                            │
│ ---                        ┆ ---                           │
│ datetime[μs]               ┆ duration[μs]                  │
╞════════════════════════════╪═══════════════════════════════╡
│ 9999-12-31 23:59:59.999999 ┆ 2932896d 22h 59m 59s 999999µs │
│ 2025-02-27 12:06:06.403591 ┆ 20146d 11h 6m 6s 403591µs     │
└────────────────────────────┴───────────────────────────────┘

In [15]: df.select(
    ...:     pl.col("dt").cast(pl.Datetime("ns"), strict=False),
    ...:     pl.col("td").cast(pl.Duration("ns"), strict=False),
    ...: )
Out[15]: 
shape: (2, 2)
┌───────────────────────────────┬─────────────────────────────────┐
│ dt                            ┆ td                              │
│ ---                           ┆ ---                             │
│ datetime[ns]                  ┆ duration[ns]                    │
╞═══════════════════════════════╪═════════════════════════════════╡
│ 1816-03-30 05:56:08.066276376 ┆ -56158d -19h -3m -51s -9337236… │
│ 2025-02-27 12:06:06.403591    ┆ 20146d 11h 6m 6s 403591µs       │
└───────────────────────────────┴─────────────────────────────────┘

In [16]: df.select(
    ...:     pl.col("dt").cast(pl.Datetime("ns"), strict=True),
    ...:     pl.col("td").cast(pl.Duration("ns"), strict=True),
    ...: )
Out[16]: 
shape: (2, 2)
┌───────────────────────────────┬─────────────────────────────────┐
│ dt                            ┆ td                              │
│ ---                           ┆ ---                             │
│ datetime[ns]                  ┆ duration[ns]                    │
╞═══════════════════════════════╪═════════════════════════════════╡
│ 1816-03-30 05:56:08.066276376 ┆ -56158d -19h -3m -51s -9337236… │
│ 2025-02-27 12:06:06.403591    ┆ 20146d 11h 6m 6s 403591µs       │
└───────────────────────────────┴─────────────────────────────────┘

In [17]: df.select(pl.col("dt").dt.cast_time_unit("ns"), pl.col("td").dt.cast_time_unit("ns"))
Out[17]: 
shape: (2, 2)
┌───────────────────────────────┬─────────────────────────────────┐
│ dt                            ┆ td                              │
│ ---                           ┆ ---                             │
│ datetime[ns]                  ┆ duration[ns]                    │
╞═══════════════════════════════╪═════════════════════════════════╡
│ 1816-03-30 05:56:08.066276376 ┆ -56158d -19h -3m -51s -9337236… │
│ 2025-02-27 12:11:25.432015    ┆ 20146d 11h 11m 25s 432015µs     │
└───────────────────────────────┴─────────────────────────────────┘

Log output

Issue description

When casting from pl.Datetime("us") to pl.Datetime("ns") (or from "ms" to "us" or "ns"), polars will not respect its own documentation about invalid values, where if strict=True the above should error and strict=False the above should return [null]. Instead it appears to overflow for values out of bounds for the more precise time_unit.

The same can be seen for pl.Duration.

Expected behavior

Polars should error, return null or wrap depending on the option passed to .cast for , rather than always wrapping, for temporal types.

Installed versions

pl.show_versions()
--------Version info---------
Polars:              1.22.0
Index type:          UInt32
Platform:            Linux-4.18.0-553.30.1.el8_10.x86_64-x86_64-with-glibc2.28
Python:              3.12.5 | packaged by conda-forge | (main, Aug  8 2024, 18:36:51) [GCC 12.4.0]
LTS CPU:             False

----Optional dependencies----
Azure CLI            <not installed>
adbc_driver_manager  1.2.0
altair               <not installed>
azure.identity       <not installed>
boto3                <not installed>
cloudpickle          2.2.1
connectorx           <not installed>
deltalake            <not installed>
fastexcel            <not installed>
fsspec               2025.2.0
gevent               <not installed>
google.auth          <not installed>
great_tables         <not installed>
matplotlib           3.8.4
numpy                1.26.4
openpyxl             <not installed>
pandas               2.1.4
pyarrow              16.1.0
pydantic             2.7.1
pyiceberg            <not installed>
sqlalchemy           1.4.49
torch                <not installed>
xlsx2csv             <not installed>