[GPU] Enable permute + onednn gemm fusion for dynamic shape

[yeonbok]

Details:

• Enable permute + onednn gemm fusion for dynamic shape
• Fix unittests to test various permute orders fused with gemm
• This pr has dependency on onednn change on memory tag. Do not merge until the memory tag change is ready

Tickets:

• 138961

[yeonbok]

LGTM. let me run daily test

Thanks, btw we need https://github.com/intel-innersource/libraries.performance.math.onednn/pull/7660 this too to run this. I will update the submodule tomorrow

=>Updated