onnxruntime

"Sticky" allocation of worker threads (#7372)

* Sticky thread alloaction

* Test sticky thread assignment

* Test sticky thread assignment

* Test sticky thread assignment

* Expose control over additional worker assignment stats

* Sticky thread alloaction

* Test sticky thread assignment

* Test sticky thread assignment

* Test sticky thread assignment

* Expose control over additional worker assignment stats

* Merge

* Merge

* Merge

* Fix Windows build

* Fix windows build 2

* Build Python 3.8 Windows CPU only

* Add env var to override binding

* Build Python 3.8 Windows CPU only

* Fix windows build

* Remove thread affinity override

* Remove goodworker

* Remove Python build settings

* Remove unneeded changes

* Remove unneeded changes

* Remove unneeded changes

* Remove unneeded changes

* Remove unneeded changes

* Remove unneeded changes

* Tidy

* Tidy

* Avoid race on preferred_worker vector

* Improve assertions

* Improve assertions

* Enum for PushBackWithTag result

* Remove unused field

* Update comments

* Extra debugging

* Extra debugging

* Extra debugging

* Support varying thread pool sizes

* Improve comments

* Remove requirement for thread local to be trivially destructible

* Use unsigned consistently for thread counts, removing casting

* Remove debug code

* Fix webassembly build

* Merge

* Merge

* Merge

* Remove unused code

* Fix build

* Extra test case for varying loop sizes

* Clean variable names

* Clean variable names

* Clean variable names

* Remove unneeded include, fix build

* Fix profiling

* Update from review comments