r2.7 cherry-pick request: Update MklMatMulPrimitiveFactory to support Arm Compute Library backend

[penpornk]

Original PR (merged into master): #52261

Add support for caching oneDNN matmul primitives in the mkl_aarch64 build. (Doesn't affect vanilla TF.) The changes in TF are small. Most of the changes are in the patch file for the oneDNN aarch64 repository.

This is more of a performance improvement (feature) PR rather than a performance regression fix. Creating a cherry-pick just in case it can be accepted.

@cfRod Could you please help post some performance numbers here (e.g., how much improvements you saw for some models)? Thank you very much!

[google-cla]

All (the pull request submitter and all commit authors) CLAs are signed, but one or more commits were authored or co-authored by someone other than the pull request submitter.

We need to confirm that all authors are ok with their commits being contributed to this project. Please have them confirm that by leaving a comment that contains only @googlebot I consent. in this pull request.

Note to project maintainer: There may be cases where the author cannot leave a comment, or the comment is not properly detected as consent. In those cases, you can manually confirm consent of the commit author(s), and set the cla label to yes (if enabled on your project).

ℹ️ Googlers: Go here for more info.

[penpornk]

Manually setting CLA to yes since the commit is from a PR that is already merged into master (#52261).

[cfRod]

Hi @penpornk, that would be approximately ~2x performance improvements for BERT benchmarks over reference oneDNN.