Shift dual stack support to use two sockets instead of mapped addresses

[divagant-martian]

Breaking changes

• Removed the listen_socket parameter from Discv5::start().
• Discv5ConfigBuilder::new() takes a ListenConfig for the sockets to listen on.
• Discv5ConfigBuilder::ip_mode() has been removed. IpMode would be determined internally according to the ListenConfig parameter.
• Discv5 and Discv5ConfigBuilder no longer implement Default since we can not determine a default value for ListenConfig.

Before

let config = Discv5ConfigBuilder::default()
    .ip_mode(discv5::IpMode::Ip4)
    .build();

let mut discv5: Discv5 = Discv5::new(enr, enr_key, config).unwrap();

let listen_socket = SocketAddr::new(Ipv4Addr::UNSPECIFIED.into(), port);

discv5.start(listen_socket).await.unwrap();

After

let listen_config = ListenConfig::Ipv4 {
    ip: Ipv4Addr::UNSPECIFIED,
    port,
};

let config = Discv5ConfigBuilder::new(listen_config).build();

let mut discv5: Discv5 = Discv5::new(enr, enr_key, config).unwrap();

discv5.start().await.unwrap();

[ackintosh]

Breaking changes

• Removed the listen_socket parameter from Discv5::start().
• Discv5ConfigBuilder::new() takes a ListenConfig for the sockets to listen on.
• Discv5ConfigBuilder::ip_mode() has been removed. IpMode would be determined internally according to the ListenConfig parameter.
• Discv5 and Discv5ConfigBuilder no longer implement Default since we can not determine a default value for ListenConfig.

Before

let config = Discv5ConfigBuilder::default()
    .ip_mode(discv5::IpMode::Ip4)
    .build();

let mut discv5: Discv5 = Discv5::new(enr, enr_key, config).unwrap();

let listen_socket = SocketAddr::new(Ipv4Addr::UNSPECIFIED.into(), port);

discv5.start(listen_socket).await.unwrap();

After

let listen_config = ListenConfig::Ipv4 {
    ip: Ipv4Addr::UNSPECIFIED,
    port,
};

let config = Discv5ConfigBuilder::new(listen_config).build();

let mut discv5: Discv5 = Discv5::new(enr, enr_key, config).unwrap();

discv5.start().await.unwrap();

[AgeManning]

This looks good to me.

There were a few things I noticed, but they seem fine to me.

As we now support ipv6, we technically shouldn't be trying to send any messages to ipv6 nodes if we are listening only on an ipv4 socket. Pretty sure this is enforced when we check if the ENR is reachable based on the ip mode.

When we send a packet we check to see if we have an open socket of the correct type of the address. If we are not, we just emit a trace error, which would be hard to find. I would suggest raising errors of this form (where we try to send an ipv6 packet but fail because we only have an ipv4 socket, or we have no socket at all). The reason I dont think this is necessary, as I think its impossible for us to do so.

It might be worth having some debug asserts or something if anyone else is worried about this. These can be put in the socket/send.rs when we try sending the packets.

One other minor suggestion could be to add some user facing helping functions around ListenConfig which is now used by all users. Maybe we could implement a Default or like an ipv4_listen() function or something. Its just ergonomics tho, and building the variant as is done in the examples seems fine to me.

Otherwise, this all looks good to me.

[AgeManning]

This looks good to me.

Ready for some merging and some thorough testing?