RandomSampleImputer rejects scalar column name 0 for observation seeds #1104

Describe the bug

RandomSampleImputer(seed="observation", random_state=0) rejects column name 0 during construction. The equivalent random_state=[0] works. Integer column names are supported, and 0 is the first column label of a pandas DataFrame created without explicit column names.

To Reproduce

import pandas as pd
from feature_engine.imputation import RandomSampleImputer

X = pd.DataFrame([[1, None], [2, 1.0], [3, 2.0], [4, None]])

# Works: column 0 supplies each observation's seed.
print(RandomSampleImputer(
    variables=[1], seed="observation", random_state=[0]
).fit_transform(X)[1].tolist())
# [2.0, 1.0, 2.0, 1.0]

# Raises ValueError before fit(), despite naming the same column.
RandomSampleImputer(variables=[1], seed="observation", random_state=0)

The constructor tests the truthiness of random_state, so it treats the integer column name 0 as missing.

Expected behavior

Both forms should select column 0 and produce the same imputation. Absent seeding variables (None or an empty list/string) should retain their existing validation.

Screenshots

Not applicable.

Desktop

• OS: Windows
• Browser: Not applicable
• Python 3.12.14, pandas 2.3.3, NumPy 2.2.6, scikit-learn 1.7.2
• Feature-engine main: d866312874a66364c6dee3c247e69476b478395c

Additional context

The existing Narwhals migration (#965) still has the same constructor guard. This report concerns the scalar integer column label on current main; it does not change the observation seed algorithm or backend support.

I am working on a small constructor validation fix with a regression comparing random_state=0 and [0]. The observation seed algorithm and backend migration remain unchanged.