feat: Add chatbot for interactive Q&A over generated documentation

[mukund-gohil-atos]

Summary

Add a conversational chatbot interface that lets users ask natural-language questions
about their repository's generated documentation, powered by the same configured LLM.

Motivation

After documentation is generated, users often have follow-up questions like:
"Where is the authentication logic?", "What does the TaskScheduler module do?",
"Show me the data flow between the API and the database layer."
A chatbot gives instant answers without manually reading through all docs.

Proposed Design

Backend

• New FastAPI endpoint POST /api/chat/{job_id} accepting { "message": "..." }
• RAG (Retrieval-Augmented Generation) pipeline:
  1. Load generated .md docs for the job
  2. Chunk & embed (using litellm embedding endpoint or local model)
  3. Vector store (in-memory with faiss or file-based with chromadb)
  4. Retrieve relevant chunks, build context prompt, call configured LLM
• Streaming response via Server-Sent Events (SSE) for real-time feel

Frontend

• Chat widget on the documentation viewer page
• Message history with user/assistant bubbles
• Markdown rendering in assistant responses (code blocks, links)
• "Clear chat" button

Configuration

• Reuse the existing LLM provider configuration (no extra setup needed)
• Optional: dedicated --chat-model config for a cheaper/faster model

Files Affected

• New codewiki/src/fe/chat_service.py — RAG logic
• New codewiki/src/fe/chat_routes.py — FastAPI chat endpoints
• codewiki/src/fe/web_app.py — register chat routes
• codewiki/src/fe/templates.py — add chat UI to doc viewer
• requirements.txt — add faiss-cpu or chromadb

Acceptance Criteria

• Users can ask questions about docs via the web UI
• Responses cite the relevant module/file
• Streaming output works (no waiting for full response)
• Works with all supported LLM providers
• Chat history persists for the session