Fix handling of None and "" as hostnames in trio.socket

[njsmith]

It looks like the Python stdlib has a weird behavior:

• socket.getaddrinfo(None, 0) resolves to the wildcard address(es)
• sock.bind((None, 0)) is an error
• socket.getaddrinfo("", 0) is an error
• sock.bind(("", 0)) binds to the wildcard address

Trio is currently not quite consistent. Trio's getaddrinfo matches the stdlib, but for bind we raise an error with both versions, because we use getaddrinfo to validate the user's input, and then if it passes we hand the user's input directly to the stdlib bind.

We should at the least support one way of calling bind, and perhaps we should teach both versions to support both behaviors.

[njsmith]

So I guess we probably shouldn't special-case "" in our getaddrinfo implementation, since it technically is a name string you could look up, and it has well-defined behavior in some cases. In particular, on Windows, it returns a list of all currently-configured local interfaces. (Why this is useful, I have no idea. But it's specified to do so.)

So I think we should:

• Support None everywhere
• Special case "" in our sockaddr resolver routines (but not getaddrinfo) by replacing it with None.

(Fun fact: on Linux getaddrinfo("*", ...) also resolves to the wildcard addresses, exactly like passing None. But this isn't portable, and None is.)

[njsmith]

There's another special case in the sockaddr resolution stuff: "<broadcast>"

[njsmith]

It looks like basically to fix this we just need to add an if address[0] == "": address[0] = None check to _resolve_local_address.

I think "<broadcast>" is not worth worrying about; it's just a hard-coded shorthand for "255.255.255.255" or INADDR_BROADCAST, and it doesn't really do anything useful these days anyway: https://stackoverflow.com/questions/683624/udp-broadcast-on-all-interfaces. Or we could do if address[0] == "<broadcast>": address[0] = "255.255.255.255".

[njsmith]

Eh, handling <broadcast> was trivial so I went ahead and did it in #383