llama.cpp adds vision support for Minimax-M3

The llama.cpp repository has merged support for the Minimax-M3 vision model, enabling local inference for the architecture.