[pa-devbot]

Adds connection_id() to AbstractServerSocket: a token identifying the connection the socket is currently serving. It increments (under the socket's lock) on every accept, and send() now takes an optional connection argument — a write that outlived its connection (the peer disconnected and a new client was accepted in the meantime) is refused instead of being delivered into the new client's socket.

• connection_id() returns the current token; send(..., connection) re-checks the token at transport selection and on every send-loop iteration (queue time and run time in the Qt backend).
• The default argument (0 = no check) keeps existing callers unchanged.

Split out of #1879 (HttpServer) at the maintainer's request: the in-app HTTP server is the motivating use case — its per-connection task may be inside a long handler call when the client disconnects, and a fast reconnect can be accepted before the task notices; without the id, its final response could be written into the new client's stream. #1879 is stacked on this PR.

Tests: new Test_Sockets_StaleSend (runs for every backend): a send for a replaced connection is refused, a send for the current connection is delivered. Full Sockets suite passes.

[pa-devbot]

AI review: Reviewed all three backends and the new test.

Correctness: POSIX/WinSocket check the id under m_lock at send() entry and re-check on every send-loop iteration (so a long write stops mid-loop if the connection is replaced); Qt also carries the id in the queued SendData and re-checks at run time. m_connection_id is incremented under the lock on every accept, and the default connection = 0 keeps existing callers' behavior unchanged. Test_Sockets_StaleSend covers exactly the motivating race (stale id refused, current id delivered) and runs for every backend.

Verified locally (aarch64): full Sockets suite passes on this branch (30/30, both POSIX and Qt); the stacked HttpServer PR (#1879, rebased onto this branch) passes its full suite (20,588 passed, 0 failed, 5 skipped).

One small doc nit below; otherwise this looks clean and ready.

[pa-devbot]

AI review: Minor: none of the backends reset m_connection_id when a connection closes, so after the first connection this returns the last id while idle, not 0 (0 only holds before the very first accept). Harmless in practice — a stale id is still refused, and when no connection is active the state check in send() refuses first — but a caller can't use connection_id() == 0 as an "idle" test. Suggest either rewording the comment ("0 before the first connection; otherwise the id of the most recently accepted connection") or actually resetting to 0 on close if the idle⇒0 guarantee is wanted.

[pa-devbot]

Fixed in c5c9e7f — with a twist: the accessor now reports 0 whenever the socket is not CONNECTED, while the stored counter stays monotonic (resetting the counter itself on close would let a queued write from the old connection match the next connection's id and slip through). send() compares against the counter, so its behavior is unchanged; Test_Sockets_StaleSend now also pins the idle⇒0 state (before the first connect and between connections). Sockets suite 30/30; the stacked #1879 (re-synced onto this commit) passes its full suite (20,588 passed, 0 failed, 5 skipped).