TE Fortunes regression on Windows

[sebastienros]

Only on Windows
Isolated to a single PR: Add three-stage enqueuer/dispatcher scheme to SocketAsyncEngine, ThreadPoolWorkQueue and ThreadPoolTypedWorkItemQueue

I see this regression was mentioned in the PR but I think it's better to create an issue to raise awareness and/or officially accepting it.

Regression in RPS

CPU has less usage (not a bad thing per se, compromise or correlation with loss of RPS)

Max TP items /s also lower

@eduardo-vp
@kouvel

[dotnet-policy-service]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

[kouvel]

The change reduces excessive parallelization, which is not a bad thing. There can be some minor tradeoffs based on timing. Excessive parallelization can be beneficial for some situations, but typically it just adds to CPU time without any other perf benefit. Maybe there are other ways to recover the throughput regression.