fix(netwatch): set SO_RECV_ANYIF on Apple UDP sockets

[1e1f]

Closes #246.

On macOS and iOS, traffic arriving on a restricted interface such as AWDL (awdl0) is delivered only to sockets that set SO_RECV_ANYIF. This sets it in SocketState::bind on Apple targets, before the configure_socket hook runs. If setsockopt fails it's logged at debug and the bind carries on as before.

HUMAN NOTE:
multi-decade programmer, writing rust since 2022 -- but this is AI assisted work. I AM NOT a sockets guru -- so this COULD break something somewhere for anybody using #[cfg(target_vendor = "apple")]. it is a NOOP on all other platforms. probably the most controversial thing is do you want that "debug!("failed to set SO_RECV_ANYIF: {:?}", err);" line or not ;)

API Changes

None.

Notes & open questions

• This adds unsafe to udp.rs: one libc::setsockopt call, plus a getsockopt in the test. As far as I can find, no safe API can set this option. libc, socket2 and nix don't define SO_RECV_ANYIF, so it's a local const (0x1104, from <sys/socket.h>). socket2's generic setsockopt is pub(crate), and nix's sockopt_impl! expands to the same unsafe block inside netwatch. Each block has a SAFETY comment. If you'd rather, I can upstream the constant to libc and a setter to socket2, so this becomes a safe call.
• test_bind_sets_recv_anyif reads the option back through the configure-socket hook, for v4 and v6. Whether AWDL then delivers can't be checked on one machine. The manual check and before/after numbers are in Apple: sockets need SO_RECV_ANYIF to receive over AWDL (and link-local addresses are never advertised) #246.
• End-to-end AWDL in iroh also needs the scope-id fix from IPv6 link-local addresses lose their scope id, so connecting to fe80::…%iface never completes iroh#4584. This PR doesn't depend on it and can merge first.
• cargo make on macOS: fmt-check, check and test (30/30) pass. clippy and doc fail on warnings in interfaces/bsd.rs, a file this PR doesn't touch: unused_async on default_route (:40) and unclosed <must …> tags in a doc comment (:519–525). clippy fails the same way on main. I've left them alone here; happy to send a separate fix if you'd like.

Change checklist

• Self-review.
• Documentation updates following the style guide, if relevant.
• Tests if relevant.
• All API changes documented.
• This PR was created by a human that thought critically about the
  proposed change and wrote an as clear and concise description as
  they could.
• This PR isn't slop, and is carefully crafted to do have the
  intented effect.
• cargo make passes locally.

[matheus23]

The only thing that was concerning me about this PR was whether enabling this unconditionally could have bad side-effects, but it turns out there's lots of precedence in enabling this in widely used apple tools:

• Apple's DCHPv4 client
• Apple's DHCPv6 client
• MacOS ping and ping6
• Another P2P networking library NewNode
• Yggdrasil from Matrix

So that looks fine.
(The reason I looked into this was because in theory you could also implement this using the configure socket hook. But obviously that then would be opt-in.)

[matheus23]

the most controversial thing is do you want that "debug!("failed to set SO_RECV_ANYIF: {:?}", err);" line or not ;)

This is the right pattern. It's also what a bunch of MacOS tools are doing, and debug is the correct level IMO.

[matheus23]

If you'd rather, I can upstream the constant to libc and a setter to socket2, so this becomes a safe call.

I think we can merge as-is, but upstreaming could be worth either way? Seems like socket2 accepts PRs for options like these.