Support for Diffusion-Augmented LLMs

[pieris98]

The idea is to implement/perform this speculative-free post-training method for popular ggml-optimized LLM models (e.g. Qwen3.8-27B).
Alternatively, when such post-trained models become available, implement support for their diffusion-augmented variants in llama.cpp / ggml.

This will possibly accelerate inference with a small memory footprint, enabling inference of popular LLM models in smaller GPUs.