LWG4476. run_loop should not have a set_error completion

[jwakely]

https://cplusplus.github.io/LWG/issue4476

This issue proposes adding noexcept to the run() and finish() members, which have narrow contracts. Violating the preconditions would be a logic error in the program; users of the run_loop scheduler should not try to run it when it's in the wrong state.

LWG would like to get LEWG confirmation that violating the Lakos Rule is appropriate here.

See also https://cplusplus.github.io/LWG/issue4215 which gives additional rationale for run_loop::finish() being non-throwing, by analogy with destructors.