fix(preprocessor): reject an unrecognized scaling value

[ChrisW09]

Summary

Preprocessor(scaling=...) accepts any string. An unrecognized value silently produces a
pipeline with no scaler at all — no error, no warning. A typo therefore disables the
scaling step rather than reporting it, while the same typo in numerical_method raises.

Reproduction

from pretab.pipeline import get_numerical_transformer_steps

for s in ["minmax", "MinMax", "zscore", "not_a_scaler", "minmaxx", "none", None]:
    print(f"{s!r:15} -> {[n for n, _ in get_numerical_transformer_steps('ple', scaling=s)]}")

'minmax'        -> ['imputer', 'minmax', 'ple']
'MinMax'        -> ['imputer', 'minmax', 'ple']
'zscore'        -> ['imputer', 'scaler', 'ple']
'not_a_scaler'  -> ['imputer', 'ple']          <- silently unscaled
'minmaxx'       -> ['imputer', 'ple']          <- silently unscaled
'none'          -> ['imputer', 'ple']          <- intentional
None            -> ['imputer', 'ple']          <- intentional

End to end, Preprocessor(scaling="minmaxx").fit(df, y) completes without complaint and
produces unscaled features.

Expected

An unrecognized scaling value raises InvalidParamError listing the valid options, exactly
as numerical_method and categorical_method already do. None and "none" keep meaning
"no scaling".

Actual

Anything that isn't a recognized scaler is treated as "no scaling".

Root cause

pretab/pipeline/numerical.py:131-135:

if scaling is not None:
    scaling = resolve_method(scaling, NUMERICAL_METHODS, NUMERICAL_ALIASES)
if scaling in scalers and scaling != method:
    steps.append(scalers[scaling])

resolve_method returns the input lowercased when it recognizes nothing, and the membership
test then just fails. Contrast with the method name a few lines below, which is explicitly
validated against NUMERICAL_METHODS and raises.

Suggested fix

Validate after resolution, allowing the two documented no-op spellings:

if scaling is not None:
    scaling = resolve_method(scaling, NUMERICAL_METHODS, NUMERICAL_ALIASES)
    if scaling not in scalers and scaling != "none":
        raise invalid_param_error(
            "get_numerical_transformer_steps", "scaling", scaling,
            "must name a scaler or disable scaling",
            valid={*scalers, "none", None},
        )

This matters more than a typical typo guard because the failure is invisible: the fit
succeeds, the widths are unchanged, and only the numeric scale of the features differs — so
the mistake surfaces (if at all) as a quietly worse model.

Environment

• pretab 0.1.0 (main @ 51c3043)
• Python 3.11.15, numpy 2.4.6, pandas 2.3.3, scikit-learn 1.9.0, scipy 1.17.1
• macOS (darwin 25.5.0)