Add an n_preprocessing_jobs option to the classifier and regressor.

[oscarkey]

And deprecate the disconnected n_jobs option. Set the default value
to 1.

Previously this option was just ignored, and 1 used in all cases. However, we've found that using higher values can be useful in some cases.

I add a new option rather than reconnecting the old option, because users might have set the old option to a large value, or -1, which can hurt performance. E.g. from some quick benchmarking on my macbook:

                                      duration_per_call_seconds
                                                     mean   std
train test cols  wrapper.n_jobs                                
500   10   10    -1                                  0.54  0.66
                  1                                  0.23  0.05
                  2                                  0.53  0.63
                  3                                  0.52  0.62
1000  10   10    -1                                  1.69  0.01
                  1                                  0.59  0.32
                  2                                  0.75  0.63
                  3                                  0.76  0.63

This way, users have to explicitly switch to the new option, and (hopefully) select a sensible value.

Also, Ruff format examples/kv_cache_fast_prediction.py.

ref RES-736

[gemini-code-assist]

Code Review

This pull request introduces a new n_preprocessing_jobs parameter to replace the deprecated and non-functional n_jobs parameter in both the classifier and regressor. This is a well-reasoned change that improves clarity and prevents potential performance issues for users. The implementation is thorough, updating function signatures, docstrings, and call sites across the codebase. I've identified a few minor issues, mainly related to copy-paste errors in documentation and warning messages, and a naming inconsistency that could be improved for maintainability. Overall, this is a solid contribution.

[oscarkey]

sorry Benjamin, this was a bit of a mess, but it should now be ready haha

[bejaeger]

Nice! LGTM.