feat: stream Reginald's replies token-by-token

The 26b is slow (~30s/call); the chat now shows the answer forming instead of
freezing until it's done. The final prose streams over SSE; tool-calling turns
stay structured (no partial tokens), so streaming kicks in for the narration.

core (@commitea/core):
- chat-client.complete gains an optional onToken — when set, it requests
  stream:true and parses the OpenAI SSE stream, emitting content deltas and
  assembling streamed tool-call argument fragments into the final result.
- GiteaHttpResponse exposes the optional `body` stream (real fetch has it; stubs
  don't). agent-loop threads onToken to each completion.

app:
- model:chat forwards each delta to the renderer (event.sender.send); preload
  exposes model.onToken(cb) → unsubscribe. useChat accumulates the live stream
  into a growing bubble (with a cursor), replaced by the authoritative final
  content when the turn resolves. Unconfigured → scripted reply, unchanged.

Verified: 118 core tests green (2 streaming: SSE content deltas + tool-call
fragment assembly), desktop typecheck clean, 14 fixture e2e green. Live: a real
turn against gemma-4-26b assembles the correct answer via the streaming path
(live-reginald green) — the reply now renders token-by-token.

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

apps/desktop/src/main/model.ts

@@ -86,11 +86,15 @@ export function registerModelIpc(): void {
     return { configured: true, model }
   })
 
-  ipcMain.handle('model:chat', async (_event, messages: ChatMessage[]) => {
+  ipcMain.handle('model:chat', async (event, messages: ChatMessage[]) => {
     if (!router) return { ok: false as const, reason: 'unconfigured' as const }
     const client = getGiteaClient()
     const model = await resolveLoadedModel(router.small.baseUrl, router.small.model)
     const chat = createChatClient({ ...router.small, model }, fetch)
+    // stream the model's prose to the renderer token-by-token
+    const onToken = (delta: string) => {
+      if (!event.sender.isDestroyed()) event.sender.send('model:chat:token', delta)
+    }
 
     // Proposals the model formulates this turn; the renderer approves them (the
     // write happens through gitea:applyChange, never inside the loop).
@@ -129,10 +133,11 @@ export function registerModelIpc(): void {
 
     try {
       const turn = await runAgentTurn({
-        complete: (m, t) => chat.complete(m, t),
+        complete: (m, t, ot) => chat.complete(m, t, ot),
         messages: [{ role: 'system', content: REGINALD_SYSTEM }, ...messages],
         tools: REGINALD_TOOLS,
         execute,
+        onToken,
       })
       return { ok: true as const, content: turn.content, steps: turn.steps, proposals }
     } catch (e) {