Relax when_all to allow other stop-token implementations than inplace_stop_token

[lewissbaker]

The specification of the when_all algorithm is currently written to require that it passes a receiver to child operations that has an environment with an inplace_stop_token as the result of the get_stop_token() query.

Given the possibility of using finite_inplace_stop_source<N> for when_all as proposed in P3409, we may want to relax the constraint on implementations of when_all to allow them to pass other types of stoppable-tokens to child operations. i.e. by not specifying what stop-token type is produced and instead just describing under what conditions a stop-request is sent to that stop-token.

The question for requirements on when_all is probably something that should be explored further in P3409, but also raises a wider question of how far do we want to constrain implementations with regards to the types they use.

[dietmarkuehl]

What is done for when_all should also be done for split: both explicitly use in_place_stop_source.

[lewissbaker]

Given that finite_inplace_stop_token was rejected in SG1, we can probably close this as "won't fix".