v2.0.0 - Community-validated Strix Halo LLM Guide

Current validated snapshot of the Strix Halo local LLM guide.\n\nHighlights:\n- Current direct Vulkan/RADV headline range: 63-97 t/s on Qwen MoE rows.\n- Qwen3-Coder 30B-A3B: 96.76 t/s on current b9049; previous b9010 peak 97.24 t/s remains historical evidence.\n- Qwen3.6 35B-A3B: 62.56 t/s balanced UD-Q4_K_M, 81.30 t/s speed-first Q4_0.\n- gpt-oss-120b MXFP4: 55.57 t/s direct llama.cpp Vulkan/RADV.\n- Independent community validation: three Corsair AI Workstation 300 systems reproduced the Qwen3-Coder path at 93.55-95.50 t/s and added power, RPC, server, and USB4 tuning data.\n- README, reproducibility notes, headline claim index, raw evidence map, social preview, and validation workflow are updated.\n\nThe live guide on main remains the source of truth for new benchmark updates.