Migrate customization points to portability macros that support dispatching to named member functions

[villevoutilainen]

Don't merge, for review only for now.

EDIT: The customization macros expand to one of three possible customization protocols:

1. tag_invoke
2. Named static member functions that take the CPO as a tag
3. Named instance member function that take the CPO as a tag (with deducing this)

[ericniebler]

I'm of half a mind to nuke execution_access. If you want to make your start() member function private, you can friend ::execution::start_t. Ditto for the other CPOs. EDIT: This is now done.

[ericniebler]

EDIT: This is done. Saving for posterity.

======================================

Next steps on this PR:

Needs a helper for defining CPOs. At first, I'll want all CPO's to look first for obj.member(tag), then Obj::member(obj, tag), and finally tag_invoke(tag, obj) and deprecate the tag_invoke option. Then I'll want to drop tag_invoke but keep the other two, and (once everybody has deducing this) look only for the obj.member(tag) form. Ideally, I would only need to change this in one place.

I'm concerned about diagnostics during the transition. They're bad enough already, I don't want to make them any worse. I'll need to play around with ways to define these CPOs before we can forge ahead porting the other CPOs to members.

But it somebody beats me to it, all the better.

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[ericniebler]

/ok to test

[ericniebler]

This effort will start again in a clean branch, since this PR is too far out of date.