Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@examples/react/chat/README.md`:
- Around line 1-6: Update the README.md for the chat example to include a short
descriptive paragraph explaining what the example demonstrates: that it
showcases end-anchored virtualization for chat-style UIs, supports prepending
history, streaming messages, and automatic follow/autoscroll behavior; place
this sentence or two near the top under the title in
examples/react/chat/README.md so developers immediately understand the key
features shown by the example.

In `@packages/virtual-core/src/index.ts`:
- Around line 646-654: findIndexByKey currently does an O(n) scan over
this.options.count using this.options.getItemKey which can be slow for very
large lists; to fix, introduce a Map<Key, number> (e.g., this.keyToIndex) that
is populated during measurement/build phases (the same place where item
sizes/indices are computed) and used by findIndexByKey to return index in O(1);
ensure the map is maintained/updated on any mutations that change indices
(prepend/insert/remove/clear) or invalidated and rebuilt when measurements are
recomputed so the mapping stays in sync with count and getItemKey.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@packages/react-virtual/e2e/app/test/chat.spec.ts`:
- Around line 53-54: Replace the brittle page.waitForTimeout(100) calls with
condition-based waits: locate the chat scroll container or message list and wait
for the expected state using Playwright's expect/locator APIs (e.g., await
expect(locator('selector-for-messages')).toHaveCount(expected) or await
expect.poll(() => locator('selector-for-scroll').evaluate(...)).toMatch(...));
in other words, swap each page.waitForTimeout(100) (and the similar occurrences
at 58-59, 78-79, 87-88) with a targeted wait that asserts the message/scroll
state change (use expect.poll for custom predicates or
locator.waitFor/expect.toHaveText/toHaveCount for standard conditions) so the
spec waits for actual UI transitions instead of a fixed sleep.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@examples/react/chat/src/main.tsx`:
- Around line 123-133: The current useLayoutEffect sets didInitialScroll and
also starts a timeout for setAutoHistoryEnabled, but React will run the cleanup
immediately when didInitialScroll changes so the timeout is cleared before
firing; fix by separating responsibilities: keep the scroll logic in the
existing React.useLayoutEffect that calls virtualizer.scrollToEnd() and
setDidInitialScroll(true) (leave that effect dependent on didInitialScroll and
virtualizer), then add a separate React.useEffect with an empty dependency array
that schedules the window.setTimeout(() => setAutoHistoryEnabled(true), 250) and
returns a cleanup that clears that timeout on unmount; alternatively, delay
calling setDidInitialScroll until after the timeout completes—use setTimeout to
call setAutoHistoryEnabled(true) then setDidInitialScroll(true) if you prefer
the single-effect approach.
- Line 2: Change the default import to the named createRoot from
'react-dom/client' and update the app bootstrap to use
createRoot(...).render(...). In the component using useLayoutEffect, prevent
didInitialScroll from being set synchronously (which cancels the pending
timeout) by moving the setDidInitialScroll call into the timeout callback after
setAutoHistoryEnabled(true) (or call setAutoHistoryEnabled(true) first, then
setDidInitialScroll(true) inside the same timeout), and ensure the cleanup still
clears the timeout; refer to createRoot, useLayoutEffect, didInitialScroll,
setAutoHistoryEnabled and setDidInitialScroll to locate the changes.

In `@packages/react-virtual/e2e/app/test/chat.spec.ts`:
- Line 53: Replace the fragile fixed sleeps (calls to page.waitForTimeout(100))
in chat.spec.ts with state-based waits: after each scroll or click, use
Playwright's expect.poll or expect.toHaveJSProperty against the chat scroll
container (e.g., check its scrollTop/scrollHeight or the relevant element's
visibility) until the expected value/visibility appears; specifically locate the
occurrences of page.waitForTimeout in chat.spec.ts and swap them for
expect.poll(() =>
document.querySelector('<scroll-selector>').scrollTop).toBeGreaterThan(...) or
expect(locator('<item-selector>')).toBeVisible() so the test waits for the
actual UI state instead of a fixed timeout.

In `@packages/virtual-core/src/index.ts`:
- Around line 1527-1536: isAtEnd()/getDistanceFromEnd() use getTotalSize() -
getSize() instead of the actual DOM max scroll offset, so a DOM scroll extent
larger than virtual size can prevent end detection; update getDistanceFromEnd to
compute distance using getMaxScrollOffset() - this.getScrollOffset() (clamped to
>=0) and keep isAtEnd(threshold = this.options.scrollEndThreshold) comparing
that result to threshold; modify references in getDistanceFromEnd, isAtEnd and
any callers relying on the old computation (functions: getDistanceFromEnd,
isAtEnd, getMaxScrollOffset, getScrollOffset, getTotalSize, getSize) to ensure
followOnAppend/end-pinned behavior works with the real max scroll offset.