fix: retry history fetch on failure + fix chunked encoding

Root cause: 1.2MB history responses sometimes failed with
ERR_CONTENT_LENGTH_MISMATCH or ERR_INCOMPLETE_CHUNKED_ENCODING,
leaving the chat empty ('Start a conversation' shown).

Fixes:
- Client retries history fetch up to 3x with backoff on failure
- Server uses res.send() instead of res.json() for large history
  payloads to avoid Content-Length calculation issues
- Only sets turns when response has actual data (prevents clearing
  existing turns on retry failure)

Author: Hex

packages/client/src/features/chat/store.ts

@@ -219,13 +219,24 @@ function connectSSE(threadKey: string): void {
   if (!threadKey) return
 
   // Fetch history via HTTP GET immediately (fast, parallel with SSE)
-  fetch(`/api/threads/${encodeURIComponent(threadKey)}/history`)
-    .then((r) => r.json())
-    .then((data) => {
-      setTurns(data.turns ?? [])
-      setHasOlderMessages(data.hasMore ?? false)
-    })
-    .catch(() => {})
+  const fetchHistory = (attempt = 0) => {
+    fetch(`/api/threads/${encodeURIComponent(threadKey)}/history`)
+      .then((r) => {
+        if (!r.ok) throw new Error(`HTTP ${r.status}`)
+        return r.json()
+      })
+      .then((data) => {
+        if (data.turns?.length) {
+          setTurns(data.turns)
+          setHasOlderMessages(data.hasMore ?? false)
+        }
+      })
+      .catch(() => {
+        // Retry up to 3 times with backoff (handles ERR_CONTENT_LENGTH_MISMATCH)
+        if (attempt < 3) setTimeout(() => fetchHistory(attempt + 1), 1000 * (attempt + 1))
+      })
+  }
+  fetchHistory()
 
   const url = `/api/threads/${encodeURIComponent(threadKey)}/events`
   eventSource = new EventSource(url)

packages/server/src/chat/routes.ts

@@ -101,14 +101,19 @@ export function createChatRoutes(chatModule: ChatModule, backend: AgentBackend,
     // Check response cache first
     const cached = historyResponseCache.get(threadKey)
     if (cached && Date.now() - cached.timestamp < HISTORY_CACHE_TTL) {
-      return res.json(cached.data)
+      const json = JSON.stringify(cached.data)
+      res.setHeader('Content-Type', 'application/json')
+      return res.send(json)
     }
 
     try {
       const result = await backend.getHistory(sessionKey)
       const data = { turns: result.turns, hasMore: result.hasMore }
       historyResponseCache.set(threadKey, { data, timestamp: Date.now() })
-      res.json(data)
+      // Use res.send to avoid Content-Length mismatch with large payloads
+      const json = JSON.stringify(data)
+      res.setHeader('Content-Type', 'application/json')
+      res.send(json)
       res.json({ turns: result.turns, hasMore: result.hasMore })
     } catch {
       res.json({ turns: [], hasMore: false })