KWin reconnect: keep fallback on dmabuf import failure and harden producer error transitions

[Copilot]

KWin could leave fallback and resume rendering after consumer reconnect even when dmabuf import failed, creating a split state (fallback=false + zero imported buffers) that can sustain high CPU. This change makes reconnect/import atomic and ensures transport failures consistently re-enter fallback.

• Reconnect transaction is now commit-on-success

  • In AnlandBackend::onReconnectTimer(), dmabuf import is attempted before publishing connected state.
  • If import fails, the candidate generation is rejected and fallback/retry remains active; notifier setup, render resume, and repaint are skipped.

• Producer-side generation rejection API

  • Added reject_consumer_resources(display_ctx*) to display_producer and used it from the backend on import failure.
  • Rejection path reuses producer fallback transition so C producer state and C++ backend state stay aligned for subsequent retries.

• Transport failure handling tightened

  • trigger_refresh() now handles sendmsg() failure/short-send by entering fallback and returning error.
  • poll_input_event(), poll_input_event_extend_fds(), and poll_input_event_extend_data() now enter fallback on receive failures / invalid framing paths that previously returned errors without guaranteed fallback transition.

if (try_exit_fallback(m_display) != 0) {
    return;
}

AnlandEglLayer *layer = m_outputs[0]->eglLayer();
if (layer && !layer->importBuffers(get_buf_count(m_display))) {
    qCWarning(KWIN_ANLAND) << "failed to import consumer buffer generation";
    reject_consumer_resources(m_display);
    return; // remain in fallback; retry later
}

// commit reconnect only after successful import
m_inFallback = false;
m_reconnectTimer->stop();
setupNotifiers();
m_outputs[0]->resumeRendering();

• Fixes KWin high-CPU loop after Anland consumer resize/reconnect when dmabuf import fails #77