[pa-devbot]

Based on the PR for the Qt ServerSocket + test harness (merge that one first, then rebase this onto main).

Adds ServerSocket_WinSocket (an accept-loop thread plus one
connection thread at a time, both from the given thread pool via plain
dispatch(), non-blocking sockets) and enables the WinSocket
instantiation of the loopback tests on Windows.

Design (mirrors the POSIX backend)

• close()/close_connection() only flip the state; the connection
  thread notices within ~10 ms and closes the socket itself
  (close_connection() also calls shutdown(SD_BOTH)), so
  close_connection() is safe to call from within a listener callback.
• The accept loop keeps running while a connection is active and
  rejects new clients via on_connect_rejected (one connection at a
  time).
• The connection task is started and published without holding the
  main lock (same AsyncTask destructor / task-slot deadlock as the
  POSIX backend).
• WSAStartup runs once via std::call_once on the thread that calls
  start(), so a ClientSocket_WinSocket created immediately after
  can use Winsock before the listener thread has run.

Windows build fix (reported by the failed Windows CI run):

• Both WinSocket headers now include <winsock2.h> instead of
  <winsock.h>: the legacy header does not declare SD_BOTH (used by
  close_connection()), which broke the build with
  error C2065: 'SD_BOTH': undeclared identifier.
• Renamed the inner non_blocking in the accept loop to
  client_non_blocking, fixing warning C4456 (it shadowed the one
  for the listen socket).

Tests: the 7 loopback scenarios now run against the WinSocket
backend on Windows in addition to Qt. The WinSocket code was
functional-tested on Linux against an in-memory mock Winsock loopback
kernel (7 scenario tests + 10-client stress, all passing).

Note (rebuild on current main): the implementation now lives in ServerSocket_WinSocket.cpp, mirroring the client header/.cpp split (#1795). Additionally, TestProgramComputer.cpp defines WIN32_LEAN_AND_MEAN before <Windows.h> so the legacy <winsock.h> is not pulled in ahead of <winsock2.h> (the two are mutually exclusive; that mix broke the Windows CI with sockaddr redefinition errors).

Update (Windows CI): the "Port in use" test failed on both Windows
CIs - the second server's bind() succeeded on a port that was
already in use. Cause: the listen socket set SO_REUSEADDR, which on
Windows (unlike POSIX) allows a second socket to bind to the same
address and port. The listen socket no longer sets it. This is now
covered by the mock Winsock kernel, which models the real Windows
SO_REUSEADDR semantics (a duplicate bind is only allowed when both
sockets set the option).

Update (formatting): fixed ServerSocket_WinSocket.cpp
indentation left over from the header/.cpp split (method bodies had
kept their in-class 8-space indent). The Mutex/ConditionVariable
wrapper migration from #1787 comes in with the rebase.