- Add POST /transcribe/sessions/{id}/export endpoint
- Generate SRT (SubRip subtitle format) with timestamps
- Generate TXT (plain text with [HH:MM:SS,mmm] timestamps)
- Generate JSON (structured data: session, segments, summaries, canvas events)
- Return as FileResponse download with Content-Disposition headers
- Filenames include sanitized session title + date
- No API keys stored or logged during export
- Create llm_client.py with 5 provider implementations (Anthropic, OpenAI, Ollama, OpenRouter, Google)
- Add build_prompt() helper to construct system/user prompts from templates
- Wire up POST /transcribe/sessions/{id}/summaries endpoint to call LLM client
- Return generated content + token counts (input_tokens, output_tokens)
- API keys passed per-request, never stored or logged
- Uses prompt templates from prompts.py based on summary_type
