datetime: propagate NaN for week and days_in_month with missing_values=ignore

[AlejandroCoronadoN]

When missing_values="ignore", DatetimeFeatures.transform() raises for the week and days_in_month features if the input has a missing date. Both cast to np.int64, which fails on the NaT-derived NA:

IntCastingNaNError: Cannot convert non-finite values (NA or inf) to integer

The other datetime features (month, year, day_of_month, ...) do not force an int cast and correctly return NaN for the missing row, so only these two were affected.

Repro:

import pandas as pd
from feature_engine.datetime import DatetimeFeatures

X = pd.DataFrame({"date": pd.to_datetime(["2022-01-01", None, "2022-03-01"])})
DatetimeFeatures(features_to_extract=["days_in_month"],
                 missing_values="ignore").fit(X).transform(X)

This casts both to float64 instead, so the missing row becomes NaN like the other features. Added a test covering week and days_in_month with missing_values="ignore".