feat(backend): add note-prompts module with Smart Actions LLM integration
- Add @bytelyst/llm dependency (file: ref) + llm.ts singleton wrapper
- Add LLM env vars to config (LLM_PROVIDER, LLM_DEFAULT_MODEL, LLM_VISION_MODEL, LLM_EMBEDDING_MODEL)
- Create note-prompts module: types, repository, runner, routes, seed (20 built-in templates)
- Built-in templates: 8 transform, 3 extract, 3 generate, 2 analyze, 2 vision, 2 export
- Prompt runner supports text, image, and text+image inputs via @bytelyst/llm vision
- Upgrade copilot-transform.ts to use @bytelyst/llm directly (with local heuristic fallback)
- Add reading-time endpoint (GET /api/notes/:id/reading-time)
- Extend agent-action types with smart_action and auto_enrich
- Add note_prompts Cosmos container to cosmos-init
- Register notePromptRoutes in server.ts
- 15 new tests (CRUD, run, slug resolution, seed validation, reading-time)
9e3a7206b9