ENH: Create logistic regression oneDAL object on the fly for predictions after fallbacks

[david-cortes-intel]

Description

This PR changes the logic of logistic regression on GPU to use oneDAL for predictions even if the model was fitted through scikit-learn, by creating a oneDAL object on-the-fly from the fitted coefficients.

---

Checklist:

Completeness and readability

• I have commented my code, particularly in hard-to-understand areas.
• Git commit message contains an appropriate signed-off-by string (see CONTRIBUTING.md for details).
• 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.

[david-cortes-intel]

/intelci: run

[david-cortes-intel]

/azp run Nightly

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[codecov]

Codecov Report

❌ Patch coverage is 16.00000% with 21 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
onedal/linear_model/logistic_regression.py	12.50%	14 Missing ⚠️
sklearnex/linear_model/logistic_regression.py	22.22%	7 Missing ⚠️

Flag	Coverage Δ
github	80.11% <16.00%> (-0.22%)	⬇️

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

Files with missing lines	Coverage Δ
sklearnex/linear_model/logistic_regression.py	45.26% <22.22%> (-0.69%)	⬇️
onedal/linear_model/logistic_regression.py	31.94% <12.50%> (-5.56%)	⬇️

🚀 New features to boost your workflow:

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

[Vika-F]

From the description it is not clear which problem does this PR solve? I.e. what's wrong with the current approach?
Is it also related to 'everything follows X' issue?

[david-cortes-intel]

From the description it is not clear which problem does this PR solve? I.e. what's wrong with the current approach? Is it also related to 'everything follows X' issue?

Scikit-learn also supports array API, so there can be cases where .fit() falls back to scikit-learn, but then .predict() still meets the conditions to offload to oneDAL, other than there not being a oneDAL object.

Apart from that, this should also make it easier later on to implement move_estimator_to.

[ethanglaser]

/intelci: run

[ethanglaser]

Looks like there are some GPU test fails to be addressed in private CI

[david-cortes-intel]

/intelci: run

[david-cortes-intel]

/intelci: run

[ethanglaser]

Looks like there is still a private CI test fail (AttributeError: 'numpy.ndarray' object has no attribute 'device') for py3.10 GPU tests - at least the error message there coincides with the changes here

[david-cortes-intel]

/intelci: run

[david-cortes-intel]

/intelci: run

[david-cortes-intel]

/intelci: run

[david-cortes-intel]

/intelci: run