Add LLaDA-7b-MoE diffusion model

[am17an]

Add support for https://huggingface.co/inclusionAI/LLaDA-MoE-7B-A1B-Instruct, MoE diffusion models similar to OLMoE (except the QK norm). Added two ggufs - bf16 and q8_0

Example command: ./llama-diffusion-cli -m llada-moe-7B-instruct-BF16.gguf -p "Write code to train MNIST in pytroch" -ngl 99 --diffusion-block-length 32 --diffusion-steps 256 -ub 256 --diffusion-algorithm 4 -fa 0 --temp 0 -sys "You are a helpful AI assistant"

[CISC]

@am17an There's a preview of LLaDa2Moe available, and it uses the same expert group selection as in BailingMoeV2, so I made it generally available in 6dd223b in case you want to take a stab at it later. :)

[am17an]

@CISC Thanks! I saw that, from what I see it looks their sampling doesn't change so it should be straightforward to add this model. Though I will wait for them to release the full version first.