In my opinion, introducing a completely new mechanism for request-response is a mistake.

I don't see any real problem we have that would necessitate this new approach.

If you want this link pairing concept, why not implement it via a dynamic-node-property that indicates the sending link with which the receiving link is associated? (Avoiding the need to wait for a server assigned address could likewise be achieved by a dynamic-node-property and advertised connection capability).

That way you have one mechanism for request-response, with the ability to modify that basic pattern for more specialised cases if necessary. Less work and less risk of increased interoperability issues.

Proposal :

To establish a request-response "pair" with a service S

Client attaches receiving link L to a dynamic source, with a dynamic-node property/value "paired-source -> S"
Client attaches sending link L to target S

The links themselves do not carry the paired property. We can still use $me as the reply-to.