[DOC] Fix misleading fit_transform docstrings

[tsbinns]

Reference issue

Example: Addresses #12766.

What does this implement/fix?

Adds custom docstrings for the fit_transform method of the CSP, SPoC, and SSD classes of the decoding module. The docstrings now accurately reflect the parameters of the fit and transform methods for each class (previously used the generic docstring of TransformerMixin).

[tsbinns]

Some tests failing for unrelated parts of the decoding module with the sklearn dev branch on Windows. Will take a look.

[tsbinns]

A couple more sklearn compliance checks are being run in the dev version than the stable v1.5, including the following which are causing problems (there might be others, these are just the first ones to fail):

• mne.decoding.LinearModel - check_dtype_object
• mne.decoding.TimeDelayingRidge - check_complex_data
• mne.decoding.ReceptiveField - check_complex_data

Is it sufficient that the failing ones are added to the list of ignored checks?

Also, when I checked the logs to see why this wasn't failing for the dev branch on Ubuntu, I see that the same tests are failing, but the operation gets cancelled before the tests can finish running and it reports a pass overall. Is this intended behaviour?

[larsoner]

Working around the problem in #12832, once that's in I'll merge main into this branch. Marking for merge-when-green, thanks @tsbinns !