[ENH] Implement a channel selection classifier wrapper

[TonyBagnall]

Describe the feature or idea you want to propose

we experimented with a wrapper for channel selection here
https://link.springer.com/chapter/10.1007/978-3-031-24378-3_9
basically use a really fast classifier, mini rocket, to score channels to help speed up a more accurate classifier, HIVE-COTE. It did not work particularly well, but I think there is milage for trying a few extensions, and we have more high dimensional TSC problems now. I will add this as a project, since the evaluation is not necessarily part of the evaluation,

Describe your proposed solution

Implement a BaseCollectionTransformer in the channel_selection package that has a classifier parameter (default to mini rocket)

1. in fit: scores on each channel using the classifiers fit_predict, then selects channels to keep based on a simple threshold (keep a proportion of channels is sufficient to start)
2. in transform: creates a new collection containing only the selected channels

Describe alternatives you've considered, if relevant

No response

Additional context

No response