add Michal's get_completion_scheduler(s) example from P2300

[atomgalaxy]

call foo(get_completion_scheduler(s), s) if valid, otherwise foo(s)"

[lewissbaker]

I think having cpo(args...) not map to tag_invoke(cpo, args...) if defined otherwise some default is probably an inconsiatent use of tag_invoke which could cause weirdness for all sorts of cpo-based wrappers.

I'll investigate to see if P2300 really needs that behaviour or whether the priority of tag_invoke overloads can be reversed.