fix(mdns): pin IPv6 multicast interface index on multi-adapter hosts

[ohrensessel]

Disclaimer

Yes, this is AI generated. I stumbled upon the problem with mDNS on Windows in a small project I was doing. Hope this is of any use.

Problem

On hosts with multiple network adapters — common on Windows with Hyper-V, WSL,
or VPN virtual interfaces — IPv6 mDNS discovery silently fails. Both the
multicast group join and the outbound multicast socket use interface index 0
(the OS default), which may resolve to a virtual adapter rather than the
physical LAN interface. Linux tolerates this because it typically has fewer
adapters and a more predictable default, but Windows does not.

The IPv4 path is unaffected because it passes the real interface address to
join_multicast_v4. The IPv6 path has two separate issues that the existing
TODO comments already acknowledge:

1. join_multicast_v6 is called with index 0 instead of the index of the
   interface that owns the bound address.
2. The send socket is built from a plain UdpSocket without setting
   IPV6_MULTICAST_IF, so outbound multicast packets exit the wrong adapter
   regardless of which interface was joined.

Fix

Resolve the interface index by matching the bound IPv6 address against the
host's network interfaces via if-addrs, then use that index for both:

• the multicast group join (join_multicast_v6)
• the outbound multicast interface (set_multicast_if_v6 on the send socket,
  built via socket2 which is already a dependency)

Falls back to index 0 when the address cannot be matched, preserving the
existing behaviour on single-interface hosts.

This closes both TODO comments left in the code for this exact issue.

Testing

Verified on a Windows host with Hyper-V and WSL virtual adapters where IPv6
mDNS previously produced no discoveries. After the fix, peers on the same
physical LAN segment are discovered correctly.