feat(assistant): стриминг ответов Квантика (фича 1/6)

Ответ модели «печатается» вживую через SSE поверх POST (fetch-stream,
не EventSource). Бэкенд: callLLMStream (stream:true, парсинг SSE upstream) +
callLLMStreamFailover (failover только до первого куска) + endpoint
POST /assistant/ask/stream (события meta|delta|done; быстрые пути FAQ/кэш/мета
отдаются одним done). buildAskMessages выделен из askModel (DRY).
Клиент: LS.assistantAskStream (fetch-stream + парсер SSE). Виджет: send()
стримит дельты как plain-текст с CSS-кареткой, на done — KaTeX-рендер,
источники, ссылки, оценка. Фоллбэк на sendNonStream (старый путь) если
стриминг недоступен/упал до первого куска. Cache-Control: no-transform
отключает буферизацию compression.

Проверено против живого шлюза: 24 дельты, первый текст ~1.3с, 100% русский.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

js/api.js

@@ -1183,7 +1183,7 @@ window.LS = {
   customSimsList, customSimGet, customSimCreate, customSimUpdate, customSimDelete,
   customSimShare, customSimClone, customSimRelated, customSimAddLink, customSimDelLink,
   gameProgressList, gameProgressSubmit,
-  assistantContext, assistantSeen, assistantDismiss, assistantSettings, assistantAsk, assistantFlashcards, assistantFeedback, assistantMemory, assistantMemoryClear, imageGen, imageGenStatus,
+  assistantContext, assistantSeen, assistantDismiss, assistantSettings, assistantAsk, assistantAskStream, assistantFlashcards, assistantFeedback, assistantMemory, assistantMemoryClear, imageGen, imageGenStatus,
   adminGetAssistant, adminSaveAssistant, adminTestAssistant, adminReindexTextbooks,
   adminSaveProvider, adminDeleteProvider, adminSetActiveProvider, adminAssistantModels,
   adminAssistantScan, adminAssistantProbe, adminAssistantApplyModels,
@@ -1423,6 +1423,41 @@ async function assistantSeen(ruleId) { return req('POST',   '/assistant/seen', {
 async function assistantDismiss(rid) { return req('POST',   '/assistant/dismiss', { ruleId: rid }); }
 async function assistantSettings(d)  { return req('PATCH',  '/assistant/settings', d); }
 async function assistantAsk(q, context, history, mode) { return req('POST', '/assistant/ask', { q, context: context || undefined, history: history || undefined, mode: mode || undefined }); }
+// Стриминговый ask: SSE поверх POST (fetch-stream). cbs: { onMeta, onDelta, onDone }.
+async function assistantAskStream(q, context, history, mode, cbs) {
+  cbs = cbs || {};
+  const token = getToken();
+  const headers = { 'Content-Type': 'application/json' };
+  if (token) headers['Authorization'] = `Bearer ${token}`;
+  const res = await fetch(API + '/assistant/ask/stream', {
+    method: 'POST', headers,
+    body: JSON.stringify({ q, context: context || undefined, history: history || undefined, mode: mode || undefined }),
+  });
+  if (!res.ok || !res.body) throw Object.assign(new Error('stream failed'), { status: res.status });
+  const reader = res.body.getReader();
+  const dec = new TextDecoder();
+  let buf = '';
+  const handle = (block) => {
+    let ev = 'message', data = '';
+    block.split('\n').forEach((ln) => {
+      if (ln.indexOf('event:') === 0) ev = ln.slice(6).trim();
+      else if (ln.indexOf('data:') === 0) data += ln.slice(5).trim();
+    });
+    if (!data) return;
+    let obj; try { obj = JSON.parse(data); } catch (e) { return; }
+    if (ev === 'delta' && cbs.onDelta) cbs.onDelta(obj.t || '');
+    else if (ev === 'meta' && cbs.onMeta) cbs.onMeta(obj);
+    else if (ev === 'done' && cbs.onDone) cbs.onDone(obj);
+  };
+  for (;;) {
+    const { value, done } = await reader.read();
+    if (done) break;
+    buf += dec.decode(value, { stream: true });
+    let idx;
+    while ((idx = buf.indexOf('\n\n')) >= 0) { const block = buf.slice(0, idx); buf = buf.slice(idx + 2); if (block.trim()) handle(block); }
+  }
+  if (buf.trim()) handle(buf);
+}
 async function assistantFlashcards(text, title, count) { return req('POST', '/assistant/flashcards', { text, title, count }); }
 async function assistantFeedback(rating, q) { return req('POST', '/assistant/feedback', { rating, q: q || undefined }); }
 async function assistantMemory()      { return req('GET',    '/assistant/memory'); }