from_pandas behaviour change/regression (series constructor "strict" changes)

[alexander-beedie]

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

FYI: @stinodego, looks related to #16939.

import pandas as pd
import numpy as np

pf = pd.DataFrame({
    "misc": np.array([
        '14.67%', '10.50%', '33.80%', date(1990, 7, 16),
    ]),
})
# misc
# 0      14.67%
# 1      10.50%
# 2      33.80%
# 3  1990-07-16

import polars as pl
pl.from_pandas(pf, schema_overrides={"misc": pl.String})
# TypeError: 'date' object cannot be converted to 'PyString'

Simplified further, the root cause looks like Series constructor strictness changes, as we get the same error even when setting strict=False:

values = np.array(['4.67%', '10.50%', '33.80%', date(1990, 7, 16)])
pl.Series("misc", values, dtype=pl.String, strict=False)
# TypeError: 'date' object cannot be converted to 'PyString'

Issue description

Previously (pre 1.0) this would convert the unexpected date value to null (as per non-strict behaviour).

Expected behavior

Previous behaviour would create the frame like so, with invalid values set to null:

shape: (4, 1)
┌────────┐
│ misc   │
│ ---    │
│ str    │
╞════════╡
│ 14.67% │
│ 10.50% │
│ 33.80% │
│ null   │
└────────┘

We should likely expose the "strict" param to from_pandas to opt-in to this behaviour (the underlying pandas_to_pyseries and pandas_to_pydf methods already accept it).

Passing through "strict" without further changes doesn't seem sufficient to fix the issue though, as demonstrated in the Series simplification.

Installed versions

Latest compiled `main`