Update Model URL

Update README.md

Update benchmarks: top 3 with competitor comparison, remove sub-90% results

Trim external benchmarks to top 4 only

Add external benchmark results (13 benchmarks, top 4 detailed)

Update README.md

Add Gold 10K benchmark results to model card

Add model card

LyraixGuard v5: Qwen3-4B SFT, 120K samples, 75/25 think split, final loss 0.4300

initial commit