DOC: Clarify that GPU package is required for array API on CPU

[david-cortes-intel]

Description

Turns out that using DPNP arrays on CPU with array API requires the DPC modules, as those objects contain sycl queues and the logic for handling them is in _onedal_py_dpc.

This PR updates the documentation to clarify that the GPU package is required for array API also on CPU.

---

Checklist:

Completeness and readability

• I have updated the documentation to reflect the changes or created a separate PR with updates 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 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.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
azure	?
github	81.95% <ø> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 25 files with indirect coverage changes

🚀 New features to boost your workflow:

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

[icfaust]

isnt pytorch cpu also array api? im worried you are conflating cpu sycl queues with array api by this PR.

[david-cortes-intel]

isnt pytorch cpu also array api? im worried you are conflating cpu sycl queues with array api by this PR.

Yes, and PyTorch is probably going to work, but not dpnp.

[icfaust]

@david-cortes-intel this note should explicitly mention dpnp cpu support because it is otherwise misleading, array api support does not require gpu support unless the data is on a oneapi device, which is the case for pytorch gpu, dpnp cpu and dpnp gpu, otherwise not. array api strict, pytorch cpu and many of the mentioned array api compat frameworks will work with a host only build

[david-cortes-intel]

@david-cortes-intel this note should explicitly mention dpnp cpu support because it is otherwise misleading, array api support does not require gpu support unless the data is on a oneapi device, which is the case for pytorch gpu, dpnp cpu and dpnp gpu, otherwise not. array api strict, pytorch cpu and many of the mentioned array api compat frameworks will work with a host only build

@icfaust There's also more to it as the upcoming sklearn1.9 will have functions like move_estimator_to which, once we support, are likely to require the GPU package, due to working with string-valued device argument which on our side we handle by creating a queue out of it and moving data, which happens in the DPC module.

[icfaust]

with that additional info that we should table or close this PR until more clarity can be added, and when sklearn 1.9 release occurs we can then update the docs here. if you want you can explicitly discuss oneapi device support, otherwise the info here is wrong

[david-cortes-intel]

with that additional info that we should table or close this PR until more clarity can be added, and when sklearn 1.9 release occurs we can then update the docs here. if you want you can explicitly discuss oneapi device support, otherwise the info here is wrong

That scikit-learn1.9 release is scheduled for mid-year, and we're aiming to be compatible with it in our 2026.1 release ahead of their release, which is the branch where this PR will go. PRs for 'move_estimator_to' support should follow soon.

[icfaust]

This PR can be tabled until those changes are made, while i wont put a changes requested on it, it does need to be changed from its current form. do realize what you are suggesting would amount to a feature loss as array api support occurs on host builds for all current sklearn array api releases (1.2-1.8). additionally this may require an internal discussion as this would be deleterious to arm and risc-v sklearnex support which require host builds (depending on its status) @syakov-intel