Guide: DeepSeek V4 Flash 0731 runs at 27+ t/s on Strix Halo

Benchmarked on a Flow Z13 with Ryzen AI MAX+ 395, Radeon 8060S, and 128GB LPDDR5X. The guide covers the Vulkan + DSpark setup, streaming 27+ t/s decode via bench-kv.sh, and documents several hardware gotchas including what doesn't work.