Running a Local LLM on AMD Radeon 780M — gfx1103, ROCm, and the GPU That Wasn't Supposed to Work 06-07