perf(desktop): reduce CPU/GPU overhead in Tauri shell

- SSH monitor: adaptive polling 2s→10s after stabilization, cheap TCP
  probe before expensive SSH subprocess check
- SSH setup: exponential backoff in wait_for_master_ready and
  wait_local_forward_ready (250ms→2s cap)
- Health checks: exponential backoff (100ms→1s / 250ms→2s) instead of
  flat intervals
- Startup recovery poll: cap at 15 retries instead of infinite
- Remove webview log target in release builds (eliminates IPC overhead)
- Set global NO_PROXY env var at startup for all loopback addresses
- Remove reqwest::blocking feature; use raw TCP for sidecar shutdown and
  SSH health checks
- Disable pinch-to-zoom on macOS via WKWebView.setAllowsMagnification
- Add WebView2 browser args on Windows (proxy bypass + disable unused
  UI features)
- Add Cargo release profile: thin LTO, codegen-units=1, strip
- Extract apply_platform_window_config for consistent window setup
- Add vibrancy toggle in Appearance settings (macOS desktop only) with
  solid background fallback when disabled

packages/ui/src/lib/device.ts

@@ -52,6 +52,10 @@ const setRootDeviceAttributes = (
 
   if (isTauriShellRuntime) {
     root.classList.add('desktop-runtime');
+    // Apply no-vibrancy class early so the first paint uses a solid background
+    // when vibrancy was previously disabled by the user.
+    const vibrancyOff = localStorage.getItem('desktopVibrancy') === 'false';
+    root.classList.toggle('no-vibrancy', vibrancyOff);
     root.style.setProperty('--is-mobile', '0');
     root.style.setProperty('--device-type', 'desktop');
     root.style.setProperty('--font-scale', '1');