WaitForVar in dependency engine not thread safe

[anirudh2290]

Related to #16431 . Please see the reproducible script here : https://github.com/anirudh2290/mxnet/blob/multithreaded_inference_poc/cpp-package/example/multithreading_engine_wait_to_read.cpp

[mxnet-label-bot]

Hey, this is the MXNet Label Bot.
Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it.
Here are my recommended label(s): Bug

[anirudh2290]

I noticed that this happens only when we call NDArray constructor in cpp pacakge and WaitToRead in main thread. If there is only waitall and not wait to read it works fine. Need to check if this is an issue with frontend CPP package.

[ddavydenko]

@mxnet-label-bot , add [Bug]

[anirudh2290]

I noticed that this happens only when we call NDArray constructor in cpp pacakge and WaitToRead in main thread. If there is only waitall and not wait to read it works fine. Need to check if this is an issue with frontend CPP package.

To verify this theory, I added another example which uses the backend NDArray data structure instead of CPP NDarray data structure and calls WaitToRead on it. This works fine (https://github.com/anirudh2290/mxnet/blob/multithreaded_inference_poc/cpp-package/example/multithreading_engine_wait_to_read2.cpp). This shows that the issue is with the NDArray data structure in CPP frontend.