fix(chat): eliminate streaming scroll jank (0.0.8)

[blove]

Summary

Real-LLM streaming exposed visible jerky scroll behavior. Profiled it: the auto-scroll-to-bottom effect was firing 37 `scrollTo()` calls in 2.87s during a single ~150-token response. 33 of those used `behavior: 'smooth'`, producing overlapping smooth-scroll animations that interrupt each other every ~80ms and never settle.

Fix

Drop `requestAnimationFrame(() => scrollTo({ behavior }))`, use direct `scrollTop = scrollHeight`. Instant, idempotent, no animation queue.

```diff

• if (isNewMessage || isNearBottom) {
• requestAnimationFrame(() => {

• el.scrollTo({ top: el.scrollHeight, behavior: isNewMessage ? 'instant' : 'smooth' });
  

• });
• }

• if (isNewMessage || isNearBottom) {
• el.scrollTop = el.scrollHeight;
• }
  ```

Same logic for the parked-reading case (user scrolled up >150px → no auto-scroll). Same logic for the new-message case (always pin to bottom). Just no animation overhead.

Verification

Profiled the same prompt before/after against a LangGraph dev backend running OpenAI gpt-4o-mini:

Metric	0.0.7	0.0.8
`scrollTo()` calls during stream	37	0
Of which `behavior: 'smooth'`	33	0
Scroll-position writes	37	1
Stream duration	2867ms	unchanged

User-visible result: streaming text settles in place without animation jitter.

Test plan

• `nx build chat` clean
• `nx test chat` clean
• `nx lint chat` clean
• Verified live with real LLM streaming (gpt-4o-mini via LangGraph)
• CI green
• Tag v0.0.8 to publish

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
cacheplane	Ready	Preview, Comment	May 2, 2026 4:58am