docs: document fit and transform on preprocessing transformers

[slegarraga]

Closes #33.

Adds NumPy-style docstrings to fit and transform on:

• CRMCleaner
• FiscalYearTransformer
• WealthPercentileTransformer

The docstrings describe frozen state, return types, and raised errors. WealthPercentileTransformer documents that percentiles are relative to the frozen training distribution in percentile_lookup_.

Verified:

• missing-docstring check prints MISSING: none then OK
• pytest --doctest-modules philanthropy passes (27)
• pytest tests/test_preprocessing.py tests/test_leakage.py passes (49)

[shivamlalakiya]

Thanks for this — the parameter/return/raises coverage on these four methods is exactly the shape the rest of the codebase uses, and calling out that percentile_lookup_ freezes the training distribution is the right thing to make explicit.

One correction before this goes in. Both CRMCleaner.transform and FiscalYearTransformer.transform document the return as:

The return type mirrors the input type.

That isn't what the code does. The branch at philanthropy/preprocessing/_transformers.py:128 (and :215) is:

if _get_pandas_output(self):
    return X_df
return X_df.to_numpy()

_get_pandas_output reads the scikit-learn set_output config, not the type of X. With the default config, passing a DataFrame in still gets an ndarray back — so the docstring would send readers the wrong way on the one thing they'd look this up for.

Suggested wording:

X_out : np.ndarray or pd.DataFrame
Cleaned feature matrix. Returns a DataFrame when the transformer is
configured with set_output(transform="pandas"), otherwise an ndarray.

Fix those two spots and I'll merge. Everything else in the PR is accurate as written.

[slegarraga]

Fixed in b9f9b3d. Both CRMCleaner.transform and FiscalYearTransformer.transform now document that the return is a DataFrame only when set_output(transform="pandas") is configured, otherwise an ndarray. Tests pass (49). Thank you for the precise review!