Basic implementation of OrdinalEncoder.

[trivialfis]

• Implement OrdinalEncoder.
• Implement dask version.
• Fix dask transformers with DataFrame input by using dask_cudf to construct return df.

Some other scikit-learn features are not available yet, for instance, encoded_missing_value, min_frequency, and max_categories.

The implementation is mostly based on the existing one hot encoder and label encoder.

I'm a bit confused by the output_type parameter and not sure how strictly it's enforced. I looked around, it seems some estimators can ignore this parameter in their returns. Would be great if there's a guideline on how to handle this parameter, along with #5645 .

Close #4456 .

[trivialfis]

I think the failure with dask is not related to this PR.

[csadorf]

I'm a bit confused by the output_type parameter and not sure how strictly it's enforced. I looked around, it seems some estimators can ignore this parameter in their returns. Would be great if there's a guideline on how to handle this parameter, along with #5645 .

I don't think the value should be ignored, but I am not sure how consistently it is being used within the dask implementation. To the best of my understanding, it is implemented correctly here.

[csadorf]

I really only found some superficial issues and would like to see one more test to explicitly test array inputs. Other than that, I ran some local tests to check whether the output_types are actually respected and it seems to work as expected. Nice job!

[trivialfis]

I think an updated cudf is causing the device ordinal error:

  - cudf                    23.12.00a140  cuda12_py310_231107_g2463b3ad53_140  rapidsai-nightly                    
  + cudf                    23.12.00a697  cuda12_py310_231115_g8a0a08f34f_697  rapidsai-nightly/linux-64        8MB
  - dask-cuda                23.12.00a26  py310_231107_ge5b240c_26             rapidsai-nightly                    
  + dask-cuda                23.12.00a28  py310_231115_gd026d6e_28             rapidsai-nightly/linux-64      189kB
  - dask-cudf               23.12.00a140  cuda12_py310_231107_g2463b3ad53_140  rapidsai-nightly                    
  + dask-cudf               23.12.00a697  cuda12_py310_231115_g8a0a08f34f_697  rapidsai-nightly/linux-64      137kB

[trivialfis]

After the first review, I picked some fixes from docformatter. However, due to the use of generated documents, it's not easy to simply apply all changes or enforce them in CI. But that could be something interesting in the long term.

[trivialfis]

This is ready for another review. :-)

[csadorf]

Thank you for addressing my comments!

[csadorf]

/merge