fix: reshape array for CascadeForestRegressor in customized mode

[xuyxu]

• fix the problem
• improve the coverage of unit tests

[xuyxu]

cc @kangwenhao.

Thanks for reporting! This is a problem due to the inconsistency on the array shape of predictors. For example, for univariate regression, the prediction of some predictors (e.g., XGBRegressor) is an array of shape (n_samples,), wheras that of other predictors could be an array of shape (n_samples, 1).

In this PR, we have added additional operations to unify the array shape under different cases.

[kangwenhao]

Thank you for your help. Good luck!

[xuyxu]

Please follow the steps below, and let me know if you have any problem:

• Uninstall the current version: pip uninstall deep-forest
• Download the wheels built in this PR: ZIP
• Select the appropriate wheel according to your platform and Python version
• Open a terminal and switch it into the folder that contains the wheel
• run pip install XXXX.whl

[xuyxu]

LGTM.