Adapt to mirage/mirage-protocols#28 changes:

[hannesm]

TCP and UDP layers (direct & socket):
type ipinput is gone
val handle no longer has a ~listener argument
val listen : t -> ~port:int -> (?keepalive) -> callback -> unit
val unlisten : t -> ~port:int -> unit

On the code level: the listener hash tables moved from stack-* to tcp and udp
(tcp/flow; udp/udp; tcp{v4,v6.v4v6}_socket; udp{v4,v6,v4v6}_socket).

This moved quite some code in the socket stack from the tcpip_stack_socket to
the respective tcp/udp layers.

Adapt to mirage/mirage-protocols#28 changes (that are required); fixes #452

Any review would be highly appreciated. The merge and release plan is to first cut a mirage-protocols release (with exising tcpip getting upper bounds), then a mirage-tcpip release with mirage-protocols lower bound.

[dinosaure]

So I did a quick look on this PR (don't have enough time to really go deeper) and I really like the interface which gives to us a chance to de-allocate the listening resource.

May be @talex5 can do an other view about lwt stuffs and the switch_off field but it seems fine for me 👍.

[hannesm]

strangely somehow the CI (analysis phase), even after "rebuild", does not seem to know anything about mirage-protocols 6.0.0 (merged yesterday ocaml/opam-repository@df836af) -- any insights how/when/why this happens?

[talex5]

We recently added signature checks for GitHub events. Possibly the opam-repository configuration needs updating with a secret. I'll see if I can find someone with access to that...

[talex5]

OK, should be building now - thanks to @dra27!

[hannesm]

CI succeeded, apart from ppc64 which timed out.