[WIP] Add NaiveEnginePerThread

[ZhennanQin]

Description

Add NaiveEnginePerThread to support multi-thread inference. To address the engine issue for #15576

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

[pengzhao-intel]

please verify the latency performance in https://github.com/apache/incubator-mxnet/tree/master/cpp-package/example/inference

[marcoabreu]

To me this feels a bit like creating an abonimation and I'm opposed towards introducing this new engine type.

NaiveEngine is afaik intended for debugging or simple straight-forward workloads. The regular threaded engine is intended for bigger (multi-threaded) workloads. You're trying to work around the fact that our APIs are not threadsafe. The root problem should be fixed instead of working around it and creating an even bigger mess.

[karan6181]

@mxnet-label-bot add [MKLDNN, Thread Safety, pr-awaiting-response]

[arcadiaphy]

Agree with @marcoabreu, we should focus on naive engine to make it more stateless instead of introducing new engines.

@ZhennanQin Thanks for your suggestions, I've worked out a way to do parallel inference now.

[pengzhao-intel]

@ZhennanQin I think we can close this PR since @arcadiaphy already have the solution and we just provide one of the possible solutions.

@arcadiaphy please verify the latency performance in https://github.com/apache/incubator-mxnet/tree/master/cpp-package/example/inference with your fix.