gcoap: fixes around empty ACKs [backport 2022.07]

[miri64]

Backport of #18429

Contribution description

This adds two fixes around the handling and sending of empty ACKs:

1. Send empty ACKs on CON response, even if there is no memo that matches this response.
2. Expire request on empty ACK when there is no response handler to wait for.

The 1. fix may be disputable. In my opinion it makes more sense to send a few short empty ACKs more than having some server sending a potentially large CON response multiple times just because the server missed a previous ACK, but I can remove this if this is for some reason not a good idea.

Testing procedure

Not sure how to test this... I saw it while running some experiments with #18386.

Issues/PRs references

None, but improves behavior in #18386.

[chrysn]

In my understanding this mainly a performance and protocol correctness (which we generally don't test extensively) fix; fine to have it backportet, but I wouldn't issue another RC for this if nothing else comes up that warrants one.

[miri64]

In my understanding this mainly a performance and protocol correctness (which we generally don't test extensively) fix; fine to have it backportet, but I wouldn't issue another RC for this if nothing else comes up that warrants one.

#18433 and #18434 might also be valid backports. But either way, yes: I don't think it makes much sense to warrant an RC for either of those. Just go into release once those backports are merged.