feat: AI Assistant with natural language app control, LiteLLM capabilities, and multimodal chat

[spelech]

Summary

This PR introduces the full In-App AI Assistant & App Copilot module with natural language app control, living prompt orchestration, dynamic LiteLLM capability discovery, local model filtering, and multimodal vision support.

Key Capabilities

1. LiteLLM Model Discovery & Local Exclusion:

   • Queries LiteLLM /model/info\ (with automatic fallbacks to /v1/model_info\ and /v1/models).
   • Extracts rich metadata (\SupportsVision, \SupportsFunctionCalling, \MaxInputTokens, \MaxOutputTokens, \SummaryBadge).
   • Strictly excludes local runtimes (\ollama/, \local/, \llama.cpp) while preserving network-hosted LiteLLM cloud models across LAN and localhost IPs.

2. Composer Model Switcher & Badges:

   • Modern capability-aware dropdown placed directly in the composer bar.
   • Shows provider tag and capability icons (👁️ Vision, ⚡ Tools, context limit pill like \1M).

3. Multimodal Vision & Clipboard Attachments:

   • File attachment button (📎) and clipboard paste (\Ctrl+V\ / \Cmd+V) for screenshots.
   • Staged attachment preview tray with thumbnail chips and \✕\ dismissal.
   • Serialized as \ImageContent\ into \Microsoft.Extensions.AI\ for full tool loop support.

4. Natural Language App Control & Living Prompts:

   • 12 native app tools via \IAiAppTools\ (VRAM telemetry, model fit, engine control, settings, and workflows).
   • Living prompts system loading markdown prompts dynamically from \Prompts/.

5. Testing & Documentation:

   • 681 unit tests passing (100% pass rate).
   • Live LiteLLM integration and E2E tests in \AiAssistantLiveLlmTests.cs.
   • Documentation in \docs/ai_assistant/\ updated per ASD-STE100 (Simplified Technical English) and formatted for VitePress.