fix(animate): serialize stagger delays across sibling blocks to prevent concurrent animation

[sleitor]

Summary

Fixes #482

When streaming markdown with multiple blocks, animations across sibling blocks revealed concurrently — a list could still be animating while the next paragraph was already fading in.

Root Cause

All blocks shared a single AnimatePlugin instance with a fixed startIndex=0. When a new block appeared, its first word started animating at delay=0ms, regardless of what was animating in the previous block.

Additionally, lastRenderNewWordCount was never written back to renderState in the rehype closure.

Fix

Introduces AnimateCursor — a lightweight shared counter ({ current: number }) that:

1. Resets to 0 at the start of each React render pass (in Streamdown)
2. Each block's AnimatePlugin reads cursor.current as its startIndex before animating
3. After animating, the plugin increments cursor.current by the number of newly animated words

This chains stagger delays automatically across all sibling blocks in render order without any manual setStartIndex wiring. Each block gets its own AnimatePlugin instance so prevContentLength tracking remains per-block independent.

Changes

• animate.ts: Add AnimateCursor type + createAnimateCursor(), add cursor option to createAnimatePlugin(), fix missing lastRenderNewWordCount write
• index.tsx: Replace single shared plugin with per-block plugins array + shared cursor; extract getBlockPlugins() helper to stay within biome complexity limits
• animate.test.ts: Add 4 cursor chaining tests (36 tests total, all passing)

[vercel]

@sleitor is attempting to deploy a commit to the Vercel Team on Vercel.

A member of the Team first needs to authorize it.