fix(chat): separate merged output segments

tests/e2e/chat-rendering-smoke.spec.ts

@@ -57,6 +57,7 @@ test('streams a run and renders markdown, code, math and mermaid without leaking
   }, sessionId)
 
   const bubble = page.locator('.message').last()
+  await expect(bubble.locator('.assistant-segment-divider')).toHaveCount(0)
   await expect(bubble.locator('strong')).toHaveText('bold')
   await expect(bubble.locator('code')).toHaveCount(2)
   await expect(bubble.locator('.katex')).toHaveCount(1)
@@ -132,11 +133,18 @@ test('groups assistant output across tool calls until the run ends', async ({ pa
   await expect(assistant).toHaveCount(1)
   await expect(assistant).toContainText('First update')
   await expect(assistant).toContainText('Second update')
+  await expect(assistant.locator('.assistant-segment-divider')).toHaveCount(1)
+  await expect(assistant.locator('.message-bubble')).toHaveCount(1)
   await expect(page.locator('.run-end')).toHaveCount(0)
   await page.evaluate((sid) => {
     (window as any).__PW_CHAT_SOCKET__.latest.__trigger('run.completed', { event: 'run.completed', session_id: sid, run_id: 'grouped-run' })
   }, sessionId)
   await expect(page.locator('.run-end')).toHaveText('Run output ended')
+  await expect.poll(async () => {
+    const bubble = await assistant.locator('.message-bubble').boundingBox()
+    const end = await page.locator('.run-end').boundingBox()
+    return bubble && end ? end.y - (bubble.y + bubble.height) : Infinity
+  }).toBeLessThan(50)
 })
 
 test('shows agent activity under its own avatar before the answer', async ({ page }) => {