Title: fix: reduce UI performance degradation in long conversations (#12053)

• Memoize findLatestSummaryIndex with useMemo to fix O(n²) per-render scan
• Extract per-message rendering into MemoizedHistoryItem (React.memo) so stable messages are skipped during streaming token updates
• Pre-compute lastUserInputIndex with useMemo keyed on history.length
• Remove per-child ResizeObserver loop in useAutoScroll; observe only container
• Wrap StepContainer and TimelineItem with React.memo

Fixes #12053

Description

In long conversations (500k+ tokens), the chat UI becomes progressively slower due to several compounding render inefficiencies:

• findLatestSummaryIndex was called inside the per-item render callback, making every full re-render O(n²) — it scanned the entire history once per message.
• No message-level memoization meant every streaming token caused all history items to re-render, even ones whose content hadn't changed.
• isLastUserInput captured the full history array in its closure and rescanned it on every call.
• useAutoScroll attached a ResizeObserver to every child DOM node — with hundreds of messages this created hundreds of observers firing on every token.

These changes fix each issue without requiring virtual scrolling, keeping the diff minimal and safe.

Checklist

• I've read the contributing guide (https://github.com/continuedev/continue/blob/main/CONTRIBUTING.md)
• The relevant docs, if any, have been updated or created
• The relevant tests, if any, have been updated or created

Screen recording or screenshot

Tested manually with a multi-turn conversation. Scrolling, streaming, and compaction display all behave correctly after the fix. A before/after profiler trace would show significantly fewer component re-renders per streaming token in long sessions.

Tests

No new tests added — the fix is purely in render/memoization logic with no changes to business logic, selectors, or reducers. Existing behaviour is preserved; only unnecessary re-renders are eliminated.

Summary by cubic

Reduce chat UI lag in long conversations by cutting O(n²) scans and avoiding unnecessary re-renders during streaming. Keeps long threads responsive; fixes #12053.

• Bug Fixes
  • Memoized findLatestSummaryIndex with useMemo to remove O(n²) per-render scans.
  • Added MemoizedHistoryItem with React.memo so stable messages don’t re-render on each token.
  • Computed lastUserInputIndex with useMemo keyed to history.length.
  • Wrapped StepContainer and TimelineItem with React.memo.
  • Simplified useAutoScroll to observe only the container instead of each child.

^{Written for commit 3351a0d. Summary will update on new commits. Review in cubic}