fix(onehot): encode categories before onehot_from_ordinal

[ChrisW09]

Fixes #17.

Problem

categorical_method="onehot_from_ordinal" appended only
OneHotFromOrdinalTransformer, which by design requires input that is already ordinal
encoded. The Preprocessor hands it raw column values, so any string column died inside
np.max(X, axis=0).astype(int):

Preprocessor(categorical_method="onehot_from_ordinal").fit(pd.DataFrame({"c": ["a","b","c"]*30}))
# ValueError: invalid literal for int() with base 10: 'c'

Nothing in that message points at the chosen method. And since _detect_column_types routes
every non-numeric column to the categorical side, this was the common case, not an edge one.

Three docs disagreed about what the method does:

source	claim
preprocessor.py:56	"integer codes then one-hot"
docs/user_guide/preprocessing.md:80	"One-hot from pre-encoded ordinals"
README.md:107	"One-hot on pre-encoded categoricals"

Fix

Insert ContinuousOrdinalTransformer ahead of OneHotFromOrdinalTransformer in the
categorical pipeline, making the behaviour match the Preprocessor docstring — which is
also the reading that makes the method usable from the Preprocessor at all.

The alternative (guard and raise, pointing users at "one-hot") would leave the method
unreachable through its own entry point, so I went with encoding.

Behaviour worth reviewing

ContinuousOrdinalTransformer numbers categories from 1 and reserves 0 for unseen
values. Two consequences, both of which I think are the right outcome but which are
user-visible:

• a column with k categories expands to k + 1 columns;
• a category unseen at fit time lands in the reserved column 0 rather than producing an
  all-zero row (which is what plain handle_unknown="ignore" one-hot does).

3 categories -> shape (90, 4)
['cat_c__c_bin_0', 'cat_c__c_bin_1', 'cat_c__c_bin_2', 'cat_c__c_bin_3']
unseen "ZZZ"  -> [1. 0. 0. 0.]

Both are now documented in the Preprocessor docstring and the user-guide table.

Tests

Four added to tests/test_categorical_pipeline.py: the reported string case, pipeline step
ordering, unseen categories landing in the reserved column, and get_feature_names_out()
width agreeing with transform.

Full suite: 484 passed, 9 xfailed. ruff check clean on changed files; pyright unchanged at
its pre-existing 71 errors.

🤖 Generated with Claude Code