n_jobs support details in docs

[Alexsandruss]

Description

Adds a doc page for n_jobs specifics of sklearnex.

---

Checklist to comply with before moving PR from draft:

PR completeness and readability

• I have reviewed my changes thoroughly before submitting this pull request.
• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation to reflect the changes or created a separate PR with update and provided its number in the description, if necessary.
• Git commit message contains an appropriate signed-off-by string (see CONTRIBUTING.md for details).
• I have added a respective label(s) to PR if I have a permission for that.
• I have resolved any merge conflicts that might occur with the base branch.

Testing

• I have run it locally and tested the changes extensively.
• All CI jobs are green or I have provided justification why they aren't.
• I have extended testing suite if new functionality was introduced in this PR.

Performance

N/A

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Flag	Coverage Δ
azure	?
github	71.96% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

see 41 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[david-cortes-intel]

We're using a different copyright header now: https://github.com/uxlfoundation/oneDAL/blob/main/CONTRIBUTING.md#license-and-copyright

[david-cortes-intel]

Suggested change

	* `n_jobs` parameter is supported for all estimators patched by |sklearnex|,
	while |sklearn| enables it for selected estimators only
	* `n_jobs` estimator parameter sets the number of threads used by the underlying |oneDAL|
	* |sklearnex| doesn't use `joblib` for parallelism in patched estimators and functions
	* The only low-level parallelism library used by |sklearnex| is oneTBB (through oneDAL)
	* The `threading` parallel backend of `joblib` is not supported by |sklearnex|
	* `n_jobs` parameter is supported for all estimators patched by |sklearnex|,
	while |sklearn| enables it for selected estimators only.
	* `n_jobs` estimator parameter sets the number of threads used by the underlying |oneDAL|.
	* |sklearnex| doesn't use `joblib` for parallelism in patched estimators and functions.
	* The only low-level parallelism library used by |sklearnex| is oneTBB (through oneDAL).
	* The `threading` parallel backend of `joblib` is not supported by |sklearnex|.

[david-cortes-intel]

Suggested change

	The only exception is multiclass LogisticRegression, which uses `joblib` for parallelism across classes.
	The only exception is multiclass LogisticRegression, which uses :external+joblib:doc:`joblib <index>` for parallelism across classes.

(perhaps it could be added in the substitutions list)

[david-cortes-intel]

There's also multi-threading from the MKL side.

[david-cortes-intel]

Suggested change

	`the calculation of the 'n_jobs' parameter value <https://scikit-learn.org/stable/glossary.html#term-n_jobs>`_.
	`the calculation of the 'n_jobs' parameter value <https://scikit-learn.org/stable/glossary.html#term-n_jobs>`__.

Link is repeated, single underscore makes it a named reference, which can cause with repetitions that change the name.

[david-cortes-intel]

Suggested change

	|sklearnex| supports the `n_jobs <https://scikit-learn.org/stable/glossary.html#term-n_jobs>`_ parameter
	of the original |sklearn| with the following differences:
	|sklearnex| supports the `n_jobs <https://scikit-learn.org/stable/glossary.html#term-n_jobs>`_ parameter
	of the original |sklearn|, with the following differences:

[david-cortes-intel]

There's also the MKL debug variable, and now the oneDAL debug variable.

[david-cortes-intel]

Suggested change

	When Scikit-learn's utilities with built-in parallelism are used (for example, `GridSearchCV` or `VotingClassifier`),
	When Scikit-learn's utilities with built-in parallelism are used (for example, :obj:`sklearn.model_selection.GridSearchCV` or :obj:`sklearn.model_selection.VotingClassifier`),

[david-cortes-intel]

I wasn't aware that such a system existed. Could you provide a link to the code where this happens?

[david-cortes-intel]

What does it mean "not supported by sklearnex"? What happens for example if you run an sklearn metaestimator (like BaggingClassifier) using that joblib backend with an sklearnex estimator inside?

[david-cortes-intel]

Thanks for adding these explanations. But it's still missing important pieces of information and leaves several questions unanswered:

• It's missing the threading part of MKL, the static linkage part, and how it interacts with environment variables, n_jobs parameter, inner_max_num_threads parameter, and threadpoolctl configurations.
• It doesn't mention how the threading works when put under a threadpoolctl context.
• The explanation is unclear about what ends up happening with the number of threads when using environment variables in addition to passing n_jobs as parameter.
• Could mention what happens with n_jobs in GPU mode.
• There's a difference in the threading configuration logic between daal4py and sklearnex, which this doc could also mention.
• It doesn't cover the part about some configurations being global, which is quite relevant when using python-based multi-threading.
• It could mention that the TBB threading doesn't automatically avoid nested parallelism when used in conjunction with OpenMP (which sklearn uses) and/or with joblib or python threads.
• Some estimators perform better when not using all threads - for example, linear regression is faster on LNL laptops when not using low-power E-cores. Perhaps could mention these sort of things here as they are relevant.

[icfaust]

@Alexsandruss make sure to merge main for latest CI checks on docs