feat(hardware): implement Can I Run It hardware compatibility and performance engine

[spelech]

Summary

This PR introduces the complete Can I Run It hardware compatibility and performance estimation suite into LocalLLMServerManager, inspired by the methodology of willitrunai.com and deeply integrated with live local NVML GPU & System RAM telemetry.

Key Capabilities

1. Accurate Mathematical Sizing Engine (ICanIRunItService / CanIRunItService):
   • LLMs: Computes exact model weight footprints across GGUF quantization levels (Q2_K to FP16), dynamic KV cache sizing by context window (2K–128K) and precision (FP16/Q8_0/Q4_0), CUDA overhead, and GPU vs. CPU layer offloading distributions with estimated tokens/sec.
   • Diffusion & Image Generation: Computes weights, CLIP/T5 text encoders, VAE decode buffers, and latent resolution scaling (512x512 to 1536x1536) for Flux.1 Dev/Schnell, SDXL, SD 3.5, and SD 1.5.
   • Video Generation: Computes DiT transformer weights, frame context buffers (49–129 frames), and VAE decode memory for Wan 2.2 14B/1.3B, LTX-Video, and HunyuanVideo.
   • Audio & 3D: Sizing and real-time factor estimations for Kokoro TTS, Faster-Whisper, XTTS-v2, MusicGen Melody, TRELLIS V2, and Hunyuan3D-2.
2. Hardware REST Minimal APIs (Endpoints/HardwareEndpoints.cs):
   • GET /api/hardware/fit: Dynamic query-string based model fit evaluation with live hardware fallback.
   • POST /api/hardware/evaluate: Polymorphic JSON payload evaluator for custom workloads.
3. Dedicated Interactive UI Tab ([⚡ Can I Run It]):
   • Live hardware telemetry banner (GPU name, VRAM total/free, RAM total/available).
   • Multi-modality switch bar (Text LLMs, Image Gen, Video Gen, Audio & Speech, 3D Gen).
   • Interactive sliders and model presets with real-time reactive recalculation.
   • Proportional stacked memory allocation bar and color-coded fit verdict card.
4. Ambient Compatibility Badges across Discovery & Library Cards:
   • Quick fit pills rendered on Hugging Face search results, CivitAI search results, and Ollama installed models with click-to-inspect routing into the Can I Run It tab.
5. Comprehensive Verification:
   • 400 total unit, integration, headless UI, and Playwright WASM browser tests passing.
   • 0 ESLint errors and 0 TypeScript compile errors.