Run threaded api-test SFTP/SCP tests on Windows

[yosuke-wolfssl]

Summary

Runs the 11 threaded echoserver+client tests in api-test (real SFTP/SCP
transport and rekey over a socket) on Windows for the first time, giving genuine
SFTP/SCP integration coverage under MSVC AddressSanitizer. Previously these tests
were stubbed out on Windows because the test harness's readiness handshake was
POSIX-only.

This is a follow-up to the windows-check.yml fix that made Windows CI actually
compile WOLFSSH_SSHD/SFTP/SCP/CERTS/X509 and run api-test + unit-test
under ASan. That restored compile and unit-level coverage; this PR adds the
integration coverage that was still missing.

Background / root cause

The threaded tests follow this shape:

InitTcpReady(&ready);
ThreadStart(echoserver_test, &ser, &serThread);  /* server thread */
WaitTcpReady(&ready);                             /* block until listening */
scp_client_connect(&ctx, &ssh, ready.port, cmd); /* connect to handed-back port */

The readiness handshake (block the client until the server is listening, then
hand back the bound port) was implemented only for POSIX threads. On MSVC
_POSIX_THREADS is undefined, so WaitTcpReady returned immediately and
SignalTcpReady never set ready->port — the client raced ahead and connected
to port 0, tripping AssertNotNull(ctx). Win32 threads themselves already worked
(ThreadStart uses _beginthreadex); only the readiness primitive was missing.

What changed

1. Win32 readiness primitive (wolfssh/test.h, examples/echoserver/echoserver.c)

Added a Win32 branch to the readiness primitive, mirroring the POSIX semantics:

• tcp_ready gains a CRITICAL_SECTION + a manual-reset event HANDLE.
• InitTcpReady / FreeTcpReady / WaitTcpReady and SignalTcpReady get
  #elif defined(USE_WINDOWS_API) branches.
• WaitTcpReady uses a bounded 30s wait instead of an indefinite one: the signal
  fires right after listen(), so the real wait is sub-second, and a server that
  dies before signaling fails the test fast instead of hanging CI. On timeout the
  port stays 0 and the existing AssertNotNull(ctx) trips.
• FreeTcpReady nulls the event handle after closing it so cleanup is idempotent.

All Win32 code is strictly additive (#elif USE_WINDOWS_API); the POSIX and
single-threaded branches are untouched, so non-Windows builds are unaffected.

2. Re-enable the stubbed tests (tests/api.c)

Dropped !defined(USE_WINDOWS_API) from the SFTP and SCP-rekey block guards so
the real test bodies compile and run on Windows. Also paired every InitTcpReady
with FreeTcpReady (api-test previously had 6 inits and 0 frees, unlike
testsuite.c/kex.c/sftp.c), releasing the new Windows event handle and
critical section as well as the pre-existing POSIX mutex/cond.

Tests that now run on Windows:

• SFTP: SendReadPacket, PartialSend, ReKey, ReKey_NonBlock, Confinement,
  SetDefaultPath
• SCP: ReKey, ReKey_NonBlock, ReKey_ToServer, ReKey_ToServer_NonBlock

3. Ephemeral port on Windows (wolfssh/test.h, tests/api.c, examples/echoserver/echoserver.c)

The tests passed -p 0 (OS-assigned ephemeral port) only on POSIX; on Windows
the server fell back to the fixed wolfSshPort because tcp_listen could not
read the bound port back. Reusing one fixed port across the serial networked
tests is the worst case for TIME_WAIT rebind on Windows, which (correctly) does
not set SO_REUSEADDR.

Let Windows use an ephemeral port like POSIX already does, so each test binds a
fresh port and never contends with a prior connection lingering in TIME_WAIT:

• test.h: drop the USE_WINDOWS_API exclusion from the getsockname port
  readback (getsockname/ntohs/socklen_t are all available on Winsock).
• tests/api.c: pass -p 0 on all platforms.
• echoserver.c: reject -p 0 only for the standalone main driver, not the
  NO_MAIN_DRIVER test harness, so the Windows api-test server accepts an
  ephemeral port.

Testing

• Non-Windows regression (macOS): the Win32 code is #elif USE_WINDOWS_API,
  so POSIX preprocessed output is unchanged. Rebuilt api-test + testsuite and
  ran under ASan+UBSan (-fsanitize=address,undefined -fno-omit-frame-pointer) —
  exit 0, no AddressSanitizer errors. (One pre-existing UBSan note at
  test.h:478, a gethostbyname misaligned char* load, is unrelated to this
  change.)
• Windows: exercised by the windows-check.yml asan-tests job — the only
  place _WIN32 + features + threads + ASan combine. The 11 tests should now run
  green under MSVC ASan.

Scope / risk

• Touches shared test infrastructure (test.h, echoserver.c) used by every
  example/test program, so the change is strictly additive
  (#elif USE_WINDOWS_API) with the POSIX/single-threaded branches unchanged.
• Net effect on non-Windows: none.
• Net effect on Windows: 11 networked tests go from stub to live SFTP/SCP
  transport + rekey coverage under AddressSanitizer.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.