Describe the workflow you want to enable

I want to be able to limit the number of output levels from the OrdinalEncoder and also have a new "unknown_value" mapped to this same encoded level. This is motivated in the following bug report

#25627

Describe your proposed solution

The functionality should work the same as in OneHotEncoder for consistency

https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.OneHotEncoder.html

There are three components:

1. max_categories parameter
2. min_frequency parameter
3. Option for ‘infrequent_if_exist’ in handle_unknown parameter

Describe alternatives you've considered, if relevant

I currently do this before the encoding but it is limited by lack of integration with handle_unknown

Additional context

No response