BUG: n_jobs=None not handled properly anymore in Parallel

[larsoner]

At least on my machine and implicitly in MNE-Python tests:

>>> import joblib
>>> with joblib.parallel_config('loky', n_jobs=2):
...     print(joblib.Parallel(n_jobs=None).n_jobs)
... 
1
>>> with joblib.parallel_config('loky', n_jobs=2):
...     print(joblib.Parallel().n_jobs)
... 
2
>>> joblib.__version__
'1.3.1'

No idea why it happens. On my machine it's from PyPI on Linux.

[larsoner]

Based on a quick read of the code I think it's due to the _Sentinel business introduced by @tomMoral in #1457

[larsoner]

... nope, actually looks like it was introduced by @jeremiedbb in #1392.

In any case, it seems like a regression that n_jobs=None is not treated the way it was before (and the same as the None that is the default).