[M5] AI Provider Abstraction Layer

[cursor]

Milestone 5 — Embedded AI via Embabel | Priority: Critical | COMPLETED in PR #27

Decoupled docgen from direct OpenAI API calls via AIProvider protocol.

• AIProvider protocol: chat(), tts(), transcribe()
• OpenAIProvider, OllamaProvider, HybridProvider implementations
• get_provider(config) factory from docgen.yaml or env var
• Refactored wizard.py, tts.py, timestamps.py
• Whisper model now configurable

Implemented in src/docgen/ai_provider.py. See issues/embedded-ai-embabel/11-ai-provider-abstraction.md.