Fixes dpctl.tensor.round on CUDA devices

[ndgrigorian]

When compiled for CUDA, std::rint would incorrectly round values halfway between two integers towards 0 (i.e., 1.5 -> 1.). The array API specification requires that these values be rounded to the nearest even integer instead.

To resolve this, std::rint has been replaced with sycl::rint, which does not rely on the current floating-point rounding mode (see SYCL specification).

As was pointed out at the time of implementation the floating-point rounding mode can vary between devices.

• Have you provided a meaningful PR description?
• Have you added a test, reproducer or referred to an issue with a reproducer?
• Have you tested your changes locally for CPU and GPU devices?
• Have you made sure that new changes do not introduce compiler warnings?
• Have you checked performance impact of proposed changes?
• If this PR is a work in progress, are you opening the PR as a draft?

[github-actions]

Deleted rendered PR docs from intelpython.github.com/dpctl, latest should be updated shortly. 🤞

[github-actions]

Array API standard conformance tests for dpctl=0.18.0dev0=py310h15de555_29 ran successfully.
Passed: 890
Failed: 11
Skipped: 91

[oleksandr-pavlyk]

It might be a good idea to scan the code base for remaining uses of std namespace transcendental functions and replace those one by one too

[coveralls]

coverage: 87.911%. remained the same
when pulling d8705a2 on fix-round-for-nvidia
into 1de00cb on master.

[ndgrigorian]

It might be a good idea to scan the code base for remaining uses of std namespace transcendental functions and replace those one by one too

I agree. I'll do this as a separate PR, but I think it's a good idea too.

[oleksandr-pavlyk]

I have opened gh-1701 for build break with LLVM SYCL compiler, it is unrelated to changes in this PR.