Add collapsible Reasoning panel for assistant messages

[LAF523]

Title

Add optional “Reasoning” panel for AI messages

Summary

This PR introduces a collapsible “Reasoning” panel for assistant messages. When a model/provider returns reasoning/thinking text (as metadata or content blocks), the UI can display it in a lightweight component without affecting the main message rendering.

What’s included

• A new Reasoning message component:
  • Collapsible panel UI aligned with existing styling tokens
  • Streaming-friendly header state with a left-to-right shimmer while content is updating
  • Allows users to pin the panel open by manually expanding it
• A small helper to extract reasoning text from common message shapes (best-effort)
• Global CSS utility for the shimmer animation
• Wiring in AssistantMessage (ai.tsx) to render the panel when reasoning is present

Demo

Component docs

<Reasoning />

Props

• content: string
  The reasoning text to display.
• defaultExpanded?: boolean (default: false)
  Whether the panel starts expanded.
• enabled?: boolean (default: true)
  Whether the panel is allowed to render at all.

Behavior

• Returns null when enabled is false or content is empty.
• Auto-expands while content is updating (streaming-like behavior).
• When updates stop, it can auto-collapse unless the user pinned it open.

Usage

const thinkingText = getThinkingText(message);

{thinkingText && (
  <Reasoning
    content={thinkingText}
    defaultExpanded={true}
    enabled={true}
  />
)}

Notes

• This feature is additive and does not change existing message rendering when reasoning is absent.

[vercel]

Someone is attempting to deploy a commit to the LangChain Team on Vercel.

A member of the Team first needs to authorize it.