MissForest ignores declared roles after column-name normalization #525

On current main, explicit simple imputation snake-matches declared target/ID names after column normalization, but the MissForest branch builds contexts that compare those declarations exactly. A custom target declared as Future Sales becomes future_sales, is classified as a numeric feature, and is fitted and filled even though the per-column rule is reported as ignored for the declared target.

Reproduction (with the ML extra installed):

import pandas as pd
import freshdata as fd

frame = pd.DataFrame({
    "Observed Hours": [float(i % 12) for i in range(80)],
    "Future Sales": [10 + 2 * (i % 12) + 0.5 * (i % 5) for i in range(80)],
})
frame.loc[frame.index[::11], "Future Sales"] = float("nan")
cleaned, report = fd.clean(
    frame, strategy="conservative", fix_dtypes=False,
    impute_strategy={"future_sales": "missforest"},
    target_column="Future Sales",
    missforest_n_estimators=10, missforest_max_iter=2,
    return_report=True, verbose=False,
)
print(cleaned["future_sales"].isna().sum())  # actual: 0; expected: 8

Using target_column="future_sales" preserves all eight gaps, as documented in the follow-up to #523. The runtime should apply the same resolved declarations already used by simple imputation to MissForest contexts, preserve declared IDs/targets, and retain the existing audited skip action. I will add regression coverage for global/per-column choices, original/normalized declarations, disabled renaming, and an eligible unprotected feature that still runs a model.