[SPARSE] Update oneMKL backends to match new sparse API

[Rbiessy]

Description

Update sparse API to follow the specification change: uxlfoundation/oneAPI-spec#522
Also see related small changes needed in the specification: uxlfoundation/oneAPI-spec#542

This also add some backend specific documentation in the docs folder. You can see how it is rendered here: https://rbiessy.github.io/oneMKL/domains/sparse_linear_algebra.html

All Submissions

• Do all unit tests pass locally? Attach a log.
• Have you formatted the code using clang-format?

[Rbiessy]

Test logs:
mklcpu_mklgpu_logs.txt

Tests are marked as skipped when some configurations are skipped but all the tests are run regardless.

[Rbiessy]

Had to force push on my branch to fix my author name and email, there was no code changes since my last comment.

[Rbiessy]

I've pushed 2f59edc to reduce the number of times get_pointer_type is called. In the other backends I had concerns this may introduce overheads for no reasons. I figured I would align the MKL backends with this.

[gajanan-choudhary]

I've finally reviewed the entire PR. This is fantastic work and top notch code. Thanks a lot @Rbiessy for being patient on the review and feedback. I have some review comments, I'm ready to approve once those are resolved.

[gajanan-choudhary]

Ready to approve after the last set of minor changes on moving around where some descriptor arguments are set and checked.

[gajanan-choudhary]

Approved pending the last 2 outstanding comments about fixing exception text. Thank you for this fantastic work and for being patient while we reviewed the PR.

[gajanan-choudhary]

Reaffirming LGTM given my last outstanding comments were resolved.