First shot at a unified proposal

[huitema]

This derives from the discussions on issue #96

[Yanmei-Liu]

Look good to me.

[qdeconinck]

I believe this PR is a good first proposal for such a unified solution. If such a solution can fit all the use cases, I'm happy to further work on this.

[huitema]

The commit a230bc1 applies the changes suggested in the previous reviews. With those changes, I think we have a good basis for discussing a unified solution in the working group.

[qdeconinck]

Some suggestions to remove NULL CID mentions, but overall this seems a good basis for further discussion.

[boucadair]

Some very minor comments.

[huitema]

Thanks @boucadair for the review. I have applied the editorial changes. You are asking for something more regarding the "specific logic" required when supporting zero-length CID. There are in fact two pieces to that: logic at the receiver, i.e., the node that chose to receive packets with zero-length CID; and logic at the sender, i.e., the node that accepts to send data on multiple paths towards a node that uses zero-length CID. This is explained in details in the section "Using Zero-Length connection ID", so I guess what we need is a reference to that section.