Add conv per-tensor input zero point.

[luweizhou2016]

trafficstars

Details:

• Enable per-tensor zp on AVX512_VNNI and AVX512_AMX ISA to use BRGCONV kernel.
• Merge the @luo-cheng2021 fix PR https://github.com/openvinotoolkit/openvino/pull/12877
• Merge optimization of convolution attribute PR to lower down the load network time latency.https://github.com/openvinotoolkit/openvino/pull/12898
• Only enable conv/group conv S8 input on AMX. Non-AMX same with 2022.1 LPT input precision config.

Tickets:

• 88228, 91221,91868

[dmitry-gorokhov]

@EgorDuplensky may I ask you to review the PR?

[EgorDuplensky]

@luweizhou2016 Could you please also write some description in PR message

[luweizhou2016]

@EgorDuplensky Review comments applied. Thx!

[luweizhou2016]

@EgorDuplensky ,Updated.

[luweizhou2016]

The current code inclue:

1. Enable per-tensor zp on AVX512_VNNI and AVX512_AMX ISA to use BRGCONV kernel.
2. Merge the @luo-cheng2021 fix PR https://github.com/openvinotoolkit/openvino/pull/12877
3. Merge optimization of convolution attribute PR to lower down the load network time latency.https://github.com/openvinotoolkit/openvino/pull/12898
4. Only enable conv/group conv S8 input on AMX. Non-AMX same with 2022.1 LPT input precision config.

[maxnick]

@EgorDuplensky , could you please take a look once again since the PR incorporated a couple of other ones?

[luweizhou2016]

@EgorDuplensky , Review comments applied. Pls help to review. Thx!

[luweizhou2016]

Thx @EgorDuplensky @maxnick . Applied reveiw comments and pls review @maxnick .

[luweizhou2016]

Thx! Applied review comments.

[maxnick]

@dmitry-gorokhov, could you please take a look?

[luweizhou2016]

Updated. Thx! @dmitry-gorokhov