recursion, inline completion, and stack usage

[ericniebler]

@lewissbaker writes about as_awaitable:

I am still a little concerned about the potential for unbounded recursion and stack-overflow if someone awaits a sender that completes synchronously in a loop with this formulation.

Maybe this should conditionally schedule itself onto a trampoline_scheduler if the operation may potentially complete synchronously (e.g. if the get_blocking() CPO from P2257 was adopted and calling it returned something other than asynchronous_completion_t).

[kirkshoop]

tail_sender can solve this better than a trampoline.