llama.cpp Adds Minimax M3 Support with Optimized CPU Prefill

The llama.cpp repository added support for the Minimax M3 model and implemented a replacement for the CPU operator during the prefill stage. This update aims to improve inference performance and latency for this specific architecture within the ggml runtime.