feat(gateway-chat): pre-chunk quoted text by sentence, play serially

Split the quoted dialogue into sentences and stream each as its own short /tts/stream
request (max_tokens 900), queued back-to-back on one shared AudioContext clock (speechHead)
so playback is gapless and in order. First sentence starts fast; each chunk is short so it
generates cleanly (no ramble/cap risk); the next sentence generates while the current plays.
A newer reply supersedes via the ttsGen counter; 🔊 replays.

tools/gateway-chat.html

@@ -147,6 +147,7 @@ function showErr(text){ const m = addMsg('error','err'); m.body.className = 'err
 // --- mOrpheus TTS: streaming-decode auto-voice of quoted text (Web Audio, low TTFA) ---
 let audioCtx = null;   // shared AudioContext
 let ttsGen = 0;        // generation counter — a newer reply's stream supersedes older ones
+let speechHead = 0;    // shared playback clock — sentences queue onto it back-to-back
 function primeAudio(){  // browsers suspend the AudioContext until a user gesture; resume on ANY
   if (!audioCtx) audioCtx = new (window.AudioContext || window.webkitAudioContext)();
   if (audioCtx.state === 'suspended') audioCtx.resume();   // interaction so async playback isn't silent
@@ -159,18 +160,20 @@ function extractQuotes(text){
   while ((m = re.exec(text)) !== null){ const q = (m[1] || m[2] || '').trim(); if (q) out.push(q); }
   return out;
 }
-async function streamSpeak(speech, tag){
-  const mine = ++ttsGen;                                // cancels any in-flight stream
-  if (!audioCtx) audioCtx = new (window.AudioContext || window.webkitAudioContext)();
-  if (audioCtx.state === 'suspended') await audioCtx.resume();
+function splitSentences(t){
+  return (t.match(/[^.!?…]+[.!?…]+["”’')\]]*\s*|[^.!?…]+$/g) || [t]).map(s => s.trim()).filter(Boolean);
+}
+// stream ONE sentence and queue its PCM onto the shared speechHead clock; resolves when the
+// sentence's audio is fully received (it plays on while the NEXT sentence starts generating).
+async function streamOne(sentence, mine){
   const url = $('ttsUrl').value.trim().replace(/\/+$/,'').replace(/\/tts$/,'') + '/tts/stream';
   const res = await fetch(url, {
     method:'POST', headers:{ 'Content-Type':'application/json' },
-    body: JSON.stringify({ text: speech, voice: ($('ttsVoice').value.trim() || 'baddy'), max_tokens: 3500 })
+    body: JSON.stringify({ text: sentence, voice: ($('ttsVoice').value.trim() || 'baddy'), max_tokens: 900 })
   });
-  if (!res.ok || !res.body){ tag.textContent = ' 🔇' + res.status; return; }
+  if (!res.ok || !res.body) return false;
   const reader = res.body.getReader();
-  let playhead = audioCtx.currentTime + 0.12, leftover = new Uint8Array(0);
+  let leftover = new Uint8Array(0);
   for(;;){
     const { done, value } = await reader.read();
     if (done || mine !== ttsGen) break;                 // stop if a newer reply took over
@@ -183,21 +186,31 @@ async function streamSpeak(speech, tag){
     for (let i = 0; i < N; i++) f32[i] = dv.getInt16(i * 2, true) / 32768;
     const abuf = audioCtx.createBuffer(1, N, 24000); abuf.getChannelData(0).set(f32);
     const src = audioCtx.createBufferSource(); src.buffer = abuf; src.connect(audioCtx.destination);
-    if (playhead < audioCtx.currentTime) playhead = audioCtx.currentTime + 0.02;  // underrun catch-up
-    src.start(playhead); playhead += abuf.duration;
+    if (speechHead < audioCtx.currentTime + 0.05) speechHead = audioCtx.currentTime + 0.05;  // underrun catch-up
+    src.start(speechHead); speechHead += abuf.duration; // queue back-to-back on the shared clock
   }
-  if (mine === ttsGen) tag.textContent = ' 🔈';
+  return true;
 }
 async function speakQuotes(text, msg){
   if (!$('ttsOn').checked) return;
   const quotes = extractQuotes(text);
   if (!quotes.length) return;
-  const speech = quotes.join('  ');
+  const sentences = splitSentences(quotes.join(' '));   // pre-chunk quoted text by sentence
   const tag = document.createElement('span'); tag.textContent = ' 🔊';
-  tag.style.cursor = 'pointer'; tag.title = 'mOrpheus — click to replay';
-  tag.onclick = () => streamSpeak(speech, tag).catch(() => tag.textContent = ' 🔇');
+  tag.style.cursor = 'pointer'; tag.title = 'mOrpheus — ' + sentences.length + ' sentence(s), click to replay';
+  const play = async () => {
+    const mine = ++ttsGen;                              // supersede any in-flight playback
+    primeAudio(); if (audioCtx.state === 'suspended') await audioCtx.resume();
+    speechHead = audioCtx.currentTime + 0.12;           // reset the shared clock for this reply
+    for (const s of sentences){                         // play sentences serially, in order
+      if (mine !== ttsGen) return;
+      if (!(await streamOne(s, mine))){ tag.textContent = ' 🔇'; return; }
+    }
+    if (mine === ttsGen) tag.textContent = ' 🔈';
+  };
+  tag.onclick = () => play().catch(() => tag.textContent = ' 🔇');
   msg.wrap.querySelector('.who').append(tag);
-  try { await streamSpeak(speech, tag); } catch (e){ tag.textContent = ' 🔇'; }
+  try { await play(); } catch (e){ tag.textContent = ' 🔇'; }
 }
 
 async function send(){