feat: add middleware to track AI coding agent traffic

[caffeinum]

Summary

Adds Next.js middleware to detect and track AI coding agents (Claude Code, Cursor, Windsurf, etc.) visiting the docs.

How it works

AI coding agents send Accept: text/markdown header when fetching documentation - browsers never do this. The middleware detects this signal and sends anonymized events to an external analytics API.

Detection signals:

• Accept: text/markdown (primary - browsers never send this)
• Accept: text/plain prioritized over text/html

Detected agents:

• Claude Code (claude, anthropic in UA)
• Cursor (cursor in UA)
• Windsurf (windsurf, codeium in UA)
• OpenCode (opencode in UA)
• Aider (aider in UA)
• Continue (continue in UA)
• GitHub Copilot (copilot, github in UA)
• Unknown AI (Accept header detected, unknown UA)

Configuration

No configuration needed. The middleware sends events to an external tracking API - no env vars or secrets required in this repo.

Data collected

• Host (docs.langfuse.com)
• Path (/docs/tracing)
• Agent type (claude-code, cursor, etc.)
• Country/city (from Vercel headers)
• Timestamp

No PII or sensitive data is collected.

Testing

# AI agent request - should show x-ai-agent header
curl -I -H "Accept: text/markdown" http://localhost:3333/docs

# Human request - no x-ai-agent header  
curl -I -H "Accept: text/html" http://localhost:3333/docs

[vercel]

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

A member of the Team first needs to authorize it.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}