Add CPOs for for_each

[msimberg]

Adds CPOs for hpx::for_each and hpx::ranges::for_each. As discussed earlier, I've tried to follow the return types in the standard as closely as possible (leaving out the function when an execution policy is used). The return types are adjusted only in the tag_invoke(for_each_t, ...) overloads (unlike what I thought initially would need to be adjusted, namely struct for_each(_n) or the foreach partitioner).

The return types are as follows:

policy	hpx::for_each	hpx::ranges::for_each	hpx::for_each_n	hpx::ranges::for_each_n
no policy	F	in_fun_result	iter	in_fun_result
non-task policy	void	iter*	iter	iter*
task policy	future<void>*	future<iter>*	future<iter>*	future<iter>*

where iter is the iterator type of the input or corresponding to the input range. * is not in the standard. We could keep iter as the return type for the non-task policy hpx::for_each, even though this is different than the standard.

The iterators were restricted by is_iterator with SFINAE, but then in the body of for_each there was a static_assert(is_input/forward_iterator). I've change this to only use is_input/forward_iterator in the template parameter list. Was the static assert there only for better error messages or also for some other reason?

[hkaiser]

The iterators were restricted by is_iterator with SFINAE, but then in the body of for_each there was a static_assert(is_input/forward_iterator). I've change this to only use is_input/forward_iterator in the template parameter list. Was the static assert there only for better error messages or also for some other reason?

These were used for better error messages.

[hkaiser]

LGTM, thanks!

[aurianer]

LGTM thanks!

[msimberg]

retest