[samuel-williams-shopify]

Split socket capabilities into individual operation traits, following the file capability model introduced in #40. Protocol code can now require SocketRead + SocketWrite for an existing connection without requiring socket creation, listening, options, or message I/O.

SocketOperation owns the shared Socket associated type. Each core operation has its own trait, and SocketScheduler is a blanket convenience aggregate. Message sending and receiving are independent capabilities, as are typed option getters and setters; SocketMessages and SocketOptions<O> automatically aggregate their respective pairs. Optional capabilities and resource registration depend only on SocketOperation.

Spell out address names throughout the public interface: SocketAddress re-exports std::net::SocketAddr, while SocketLocalAddress::socket_local_address and SocketPeerAddress::socket_peer_address replace the abbreviated inspection names. Both the executor and facade export the address type; standard-library addresses remain directly interchangeable.

Migrate native readiness, io_uring, Tokio, and simulated implementations while preserving operation signatures and runtime behavior. Update examples and conformance helpers to declare the operations they use, add a minimal in-memory read/write implementation and compile-time capability examples, and document migration for implementors and callers. This is a source-breaking API refactor: concrete callers import individual operation traits, and qualified resource paths use <S as SocketOperation>::Socket.