Enhance video handling and performance optimizations

- Added environment variables to prevent CPU thread pools from busy-waiting.
- Deferred loading of video models until first use to reduce VRAM footprint.
- Implemented streaming of speaking clips for improved responsiveness.
- Introduced a queue for managing speaking clips to handle multiple requests smoothly.
- Updated video playback logic to ensure proper handling of clip generation.

static/app.js

@@ -24,6 +24,8 @@ let videoModeName = "off";      // "off" | "library" | "reflective"
 let idleClipUrl = null;         // URL string (server-served) or null
 let pendingSpeakingClipMeta = null; // {chunk_id, duration_ms, text} waiting for MP4 binary
 let currentSpeakingClipBlobUrl = null;
+let speakingClipQueue = [];     // [{blobUrl, meta}] clips waiting to play
+let currentClipGeneration = 0; // incremented each clip start; guards stale onended handlers
 
 const chatArea = document.getElementById("chat-area");
 const statusBadge = document.getElementById("status-badge");
@@ -131,67 +133,86 @@ function refreshStage() {
   if (videoModeEnabled && idleClipUrl) {
     stageEl.classList.add("active");
     if (avatarVideo.src !== location.origin + idleClipUrl) {
-      avatarVideo.src = idleClipUrl;
-      avatarVideo.loop = true;
-      avatarVideo.muted = true;
-      avatarVideo.play().catch(() => {});
+      _returnToIdle();
     }
   } else {
     stageEl.classList.remove("active");
   }
 }
 
+function _returnToIdle() {
+  if (!idleClipUrl) return;
+  avatarVideo.onended = null;
+  avatarVideo.loop = false;
+  avatarVideo.muted = true;
+  avatarVideo.src = idleClipUrl;
+  avatarVideo.play().catch(() => {});
+}
+
 function playSpeakingClip(arrayBuffer, meta) {
-  // Replace the idle loop with the speaking clip.
-  stopSpeakingClip();
   const blob = new Blob([arrayBuffer], { type: "video/mp4" });
-  currentSpeakingClipBlobUrl = URL.createObjectURL(blob);
+  const blobUrl = URL.createObjectURL(blob);
+
+  if (currentSpeakingClipBlobUrl !== null) {
+    // A clip is already playing — queue this one.
+    speakingClipQueue.push({ blobUrl, meta });
+  } else {
+    _startSpeakingClip(blobUrl, meta);
+  }
+}
+
+function _startSpeakingClip(blobUrl, meta) {
+  const gen = ++currentClipGeneration;
+
+  if (currentSpeakingClipBlobUrl) {
+    URL.revokeObjectURL(currentSpeakingClipBlobUrl);
+  }
+  currentSpeakingClipBlobUrl = blobUrl;
 
   avatarVideo.loop = false;
   avatarVideo.muted = false;
-  avatarVideo.src = currentSpeakingClipBlobUrl;
+  avatarVideo.src = blobUrl;
 
-  // Show the full reply text now — the MP4 plays it in one shot so there's
-  // no per-chunk sync to do.
   if (meta && meta.text) {
     appendAssistantText(meta.text);
   }
   isPlaying = true;
 
   avatarVideo.onended = () => {
-    isPlaying = false;
-    finalizeAssistantMessage(false);
-    // Return to idle loop.
-    if (idleClipUrl) {
-      avatarVideo.loop = true;
-      avatarVideo.muted = true;
-      avatarVideo.src = idleClipUrl;
-      avatarVideo.play().catch(() => {});
-    }
-    if (currentSpeakingClipBlobUrl) {
-      URL.revokeObjectURL(currentSpeakingClipBlobUrl);
-      currentSpeakingClipBlobUrl = null;
+    if (currentClipGeneration !== gen) return; // stale handler from a replaced clip
+    URL.revokeObjectURL(currentSpeakingClipBlobUrl);
+    currentSpeakingClipBlobUrl = null;
+
+    const next = speakingClipQueue.shift();
+    if (next) {
+      _startSpeakingClip(next.blobUrl, next.meta);
+    } else {
+      isPlaying = false;
+      finalizeAssistantMessage(false);
+      _returnToIdle();
     }
   };
+
   avatarVideo.play().catch((e) => {
     console.error("speaking clip play failed:", e);
   });
 }
 
 function stopSpeakingClip() {
+  // Discard any queued clips.
+  for (const { blobUrl } of speakingClipQueue) {
+    URL.revokeObjectURL(blobUrl);
+  }
+  speakingClipQueue = [];
+  currentClipGeneration++; // invalidate any in-flight onended handlers
+
   if (!currentSpeakingClipBlobUrl) return;
-  try {
-    avatarVideo.pause();
-  } catch (_) {}
+  try { avatarVideo.pause(); } catch (_) {}
+  avatarVideo.onended = null;
   URL.revokeObjectURL(currentSpeakingClipBlobUrl);
   currentSpeakingClipBlobUrl = null;
-  if (idleClipUrl) {
-    avatarVideo.loop = true;
-    avatarVideo.muted = true;
-    avatarVideo.src = idleClipUrl;
-    avatarVideo.play().catch(() => {});
-  }
   isPlaying = false;
+  _returnToIdle();
 }
 
 async function uploadAvatar() {