change the adaptors to preserve the "non-dependently typed" property

[ericniebler]

If a sender declares its completion signatures with a nested completion_signatures alias, then it is not a dependently-typed sender; that is, we don't need an environment to know how it completes.

Currently, the adaptors do not preserve this property, but they could and arguably should.

The algorithms can use that information to type-check non-dependently-typed senders early, which would be a huge usability improvement.

[villevoutilainen]

YES, PLEASE. This allows diagnosing easy mistakes in common attempts at sender-continuation connections near that connection, not far away where the pipeline gets launched. This is essential.

[lewissbaker]

I believe that this should be addressed by P3164R0 when that paper is eventually accepted.

See also related issue #206