[FORK][FEATURE][x64]Extend InnerProduct source dynamic quantization to BF16

[liubo-intel]

Description

Enables the BRGEMM InnerProduct dynamic-quantization path (previously F32-only) to accept BF16 sources, which is required by the OpenVINO CPU plugin's BF16 compressed-FC flow.

[maxnick]

1. Please squash these commits into one
2. If these changes are original in our for (not something that we cherry-picked from oneDNN main), please use the agreed naming convention (e.g. "[FORK][FEATURE][x64] Extend inner product src dynamic quant to bf16").

[liubo-intel]

• Please squash these commits into one
• If these changes are original in our for (not something that we cherry-picked from oneDNN main), please use the agreed naming convention (e.g. "[FORK][FEATURE][x64] Extend inner product src dynamic quant to bf16").

Hi, @maxnick : I have squashed the two commits into one.
For the second point, my understanding is that you meant updating the PR title? If the [FORK][FEATURE][x64] prefix is also applied to the commit subject, it triggers the PR CI commit-format check failure. Please correct me if I misunderstood.

[maxnick]

• Please squash these commits into one
• If these changes are original in our for (not something that we cherry-picked from oneDNN main), please use the agreed naming convention (e.g. "[FORK][FEATURE][x64] Extend inner product src dynamic quant to bf16").

Hi, @maxnick : I have squashed the two commits into one. For the second point, my understanding is that you meant updating the PR title? If the [FORK][FEATURE][x64] prefix is also applied to the commit subject, it triggers the PR CI commit-format check failure. Please correct me if I misunderstood.

Hi. Got it. This naming convection doesn't suite the CI check. So what about cpu: fork: x64: extend inner product src dynamic quant to bf16? This should pass checks and still it's better to clearly mark modifications that are made only in the fork.