UI improvements: support for more data types, auto-refresh

[pamelafox]

This pull request introduces several improvements to the Agent Inbox:

• Auto-refresh of inbox when new thread exists
• Handling of more data types for the form: booleans (radio), null (empty text), and lists of strings (a fancy component which acts like the "labels" editor on GitHub)
• Reset the reset-CSS for the right-hand side description, so that the descriptions can render better with richer markdown (bullets, tables, etc). The tailwind reset CSS was making the descriptions hard to read.

I realize this is a lot of improvements at once - once I started customizing, I didn't stop! But I can separate the PR into multiple, or only send PRs for certain features, whatever you'd prefer.

Screenshot of form with more data types:

Here's the AI-generated breakdown of the changes:

UI/UX Improvements:

• Restored default spacing and list styles for headings, paragraphs, lists, and blockquotes in globals.css, softening Tailwind's Preflight reset for better readability. Table styling is also standardized with global borders and background for headers. [1] [2]

Input Handling Enhancements:

• Added support for editing boolean and string array fields in the inbox item input: booleans now render as radio buttons, and string arrays use the new TagInput component. The input update logic was refactored to handle single values, arrays, and batch updates robustly. [1] [2] [3] [4]
• Switched textareas to use autosizing (react-textarea-autosize), improving vertical sizing and user experience for longer responses. [1] [2] [3]

Agent Inbox Polling and Settings:

• Implemented auto-refresh (polling) for inbox threads, with a new autoRefreshEnabled state and a toggle in the settings popover. The polling interval is set to 15 seconds and respects the loading state to avoid overlapping fetches. [1] [2] [3] [4] [5] [6] [7]

Documentation:

• Updated the README.md to include instructions for starting the web server with yarn dev.

[vercel]

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

A member of the Team first needs to authorize it.

[pamelafox]

I sent this in a separate PR earlier