Retry listener select() on EINTR instead of exiting the listener thread

[dmitriy-bty]

Summary

The background listener thread (listenerRun in src/libvncserver/main.c),
used when the server runs with rfbRunEventLoop(screen, usec, TRUE), exits
permanently the first time select() fails with EINTR. Once it exits, the
listening sockets stay open but nothing ever calls accept(), so new clients
complete the TCP handshake and then hang forever. This fixes #732.

Root cause

listenerRun treats any select() error as fatal:

if (select(screen->maxFd+1, &listen_fds, NULL, NULL, &tv) == -1) {
    rfbLogPerror("listenerRun: error in select");
    return THREAD_ROUTINE_RETURN_VALUE;   /* thread ends here, for good */
}

EINTR is not a real failure — it just means a signal was delivered while the
thread was blocked in select(). Per signal(7), select() is one of the
calls that is not automatically restarted by the kernel after a signal
handler runs, even when the handler is installed with SA_RESTART. So a single
stray signal landing on the listener thread makes select() return -1 with
errno == EINTR, the thread returns, and the accept loop is gone. The
listening file descriptors are never closed, so from a client's point of view
the port is open and connect() succeeds, but the connection is never accepted
and the viewer waits indefinitely.

Change

Guard the error path with an EINTR check: on an interrupted select(),
continue the loop and re-enter select() instead of logging an error and
returning. All other errno values keep the existing log-and-return behavior,
so genuine failures are still surfaced exactly as before.

if (select(screen->maxFd+1, &listen_fds, NULL, NULL, &tv) == -1) {
    if (errno == EINTR)
        /* A signal interrupted select(); this is benign ... retry. */
        continue;
    rfbLogPerror("listenerRun: error in select");
    return THREAD_ROUTINE_RETURN_VALUE;
}

This is the standard EINTR-retry idiom. errno.h is already included in this
file, and the change is scoped to the single reported site in
src/libvncserver/main.c.

Validation

• Builds cleanly with the CMake build (cmake -S . -B build && cmake --build build --target vncserver), pthreads threading backend, no new warnings.
• Runtime reproducer (RED → GREEN). A small harness starts a server with
  rfbRunEventLoop(screen, usec, TRUE), lets the listener thread reach
  select(), then delivers SIGUSR1 to that thread via pthread_kill() to
  force EINTR, and finally opens a TCP connection and waits for the RFB
  ProtocolVersion greeting the server sends on accept():
  • Before the fix: listenerRun: error in select: Interrupted system call is
    logged, the listener thread dies, no greeting arrives, the client read
    times out.
  • After the fix: the listener survives the interruption, accepts the client
    and sends RFB 003.008; the connection succeeds.

Related (not changed here)

The per-client processing loop in clientInput (same file) has the same shape —
select() returning < 0 breaks the loop, terminating that client's thread,
without an EINTR guard. That path only tears down one already-connected
client rather than the whole accept loop, so it is a separate, lower-severity
issue; I have kept this PR scoped to the listener thread reported in #732 and
am happy to follow up on the client loop separately if you'd like.

Closes #732

[bk138]

#734 was more complete, merged that one.