Add KAI implementation guide for AI agent context

[ottomansky]

Summary

• Adds KAI_IMPLEMENTATION_GUIDE.md — a comprehensive reference document extracted from the profitline-js-app fi-demo branch
• Covers the full KAI integration architecture: polling proxy pattern, backend (FastAPI) and frontend (React/Next.js) implementation, chat message protocol, UI patterns, and file reference map
• Intended as system prompt/context for AI agents implementing KAI in Keboola Data Apps

Contents

1. Architecture overview (three-tier deployment, polling proxy pattern, service discovery)
2. Environment & configuration (env vars, Nginx, Next.js rewrites)
3. Backend implementation (FastAPI endpoints, stream buffering, auth)
4. Frontend implementation (React Context, polling client, system context injection, conversation storage)
5. Chat message protocol (message formats, SSE event types, tool approval)
6. Key code patterns (3 minimal, reusable snippets)
7. UI patterns (floating widget, pin-to-dashboard, suggestions, markdown rendering)
8. File reference map & design decisions table

Test plan

• Review document for accuracy against profitline-js-app fi-demo branch
• Verify code snippets compile/run correctly when used as reference
• Confirm no sensitive tokens or credentials are included

🤖 Generated with Claude Code

[jordanrburger]

ngl, seems like a lot. Also, it will go stale quickly if it's just referencing some branch in another repo. I'd say to extract out the branch of profitline-js-app into a new repo and consider it a "reference" python JS app that can be included in the docs too.