fix: disable browser DSP on screen share audio and harden ICE stats resolution

Screen share audio was muffled/gated because getDisplayMedia used plain
`audio: true`, letting the browser apply voice-optimized DSP (NS, AEC, AGC)
to desktop audio. Now passes explicit constraints disabling all processing
and requesting stereo capture.

Also improves WebRTC stats: three-tier ICE candidate-pair discovery
(transport → active-bytes heuristic → legacy fallback), height-based
simulcast layer inference, and documents mDNS obfuscation limitation.

packages/web/src/utils/screenShare.ts

@@ -121,7 +121,12 @@ export async function startScreenShare(room: Room): Promise<boolean> {
 
   try {
     const track = await room.localParticipant.setScreenShareEnabled(true, {
-      audio: true,
+      audio: {
+        echoCancellation: false,
+        noiseSuppression: false,
+        autoGainControl: false,
+        channelCount: 2,
+      },
       resolution: { width: opts.capture.width, height: opts.capture.height },
       // @ts-ignore — LiveKit accepts frameRate at capture level
       frameRate: opts.capture.frameRate,