Investigate ppltasks dependency for vNext

[StephanTLavavej]

I'm going to need to revert #2654 and #3255, so here are vNext todos that were mentioned in the latter PR's added comment:

When we can break ABI, we should:

• In <ppltasks.h>, reconsider whether _Task_impl_base::_Wait() should throw invalid_operation; it's questionable whether that's conforming, and if users want to block their UI threads, we should let them.
• Investigate whether we can avoid the ppltasks dependency entirely, making all of these issues irrelevant.

[AlexGuteniev]

• Investigate whether we can avoid the ppltasks dependency entirely, making all of these issues irrelevant.

I think that <ppltasks.h> is only used to support asynchronous (non-deferred) std::async, to spawn separate workers.

Instead of using <ppltasks.h> for that, we should use thread pool directly, like we currently do in <execution> / parallel_algorithms.cpp.

[AlexGuteniev]

Related #2206 (comment)

<future> includes <ppltasks.h> and thus drags a bunch of (more than 60) non-Standard non-reserved identifiers in.