llama.cpp b10199 adds inp embd support for next token generation

The latest release of llama.cpp (b10199) introduces significant server-side enhancements including the ability to generate the next token using input embeddings and sampled tokens. This update enables more sophisticated control over text generation processes, allowing operators to fine-tune model outputs with custom embeddings directly. Additionally, it includes bug fixes that improve stability and compatibility across different platforms, ensuring smoother operation on consumer hardware such as Apple Silicon Macs and RTX GPUs. Operators running local AI workloads should consider upgrading to leverage these new features.