[CPU][oneDNN] cherry-pick onednn reorder fix

[azhai219]

Details:

• replace onednn revert fix with root fix
• master has been fixed by reverting. Now, onednn fixed the bug. Let's evaluate whether we need to drop the previous revert and then cherry pick their fix.
• "onednn pr: cherry pick reorder oneDNN#309"

Tickets:

• 182686

AI Assistance:

• AI assistance used: no / yes
• If yes, summarize how AI was used and what human validation was performed (build/tests/manual checks).

[rkazants]

any test on our side what does it fix?

[azhai219]

any test on our side what does it fix?

It is an onednn kernel selection case. It doesn't make sense of binding some oneDNN verbose because they are always optimizing the kernel. Maybe they can provide better kernel in the next round. If we bind kernel, we may not benefit from them.

[rkazants]

any test on our side what does it fix?

It is an onednn kernel selection case. It doesn't make sense of binding some oneDNN verbose because they are always optimizing the kernel. Maybe they can provide better kernel in the next round. If we bind kernel, we may not benefit from them.

I did not get. If this is functional fix, we should cover this fix by a new test.

[vpirogov]

@rkazants, this PR picks up a proper fix for MFDNN-14804 from oneDNN upstream reducing the delta in the fork. It should not affect OpenVINO functionality or performance.

[azhai219]

The fix has been verified. Although previous revert #34872 has been merged, we won't merge it.
It won't be an issue in the following integration because onednn fix was merged into main.
close it.