[KERNEL] AllSpark + Exllama vLLM

[Qubitium]

Let's port these 2 kernels over to GPTQModel as well for simple inference.

• AllSpark by @wyajieha: The difference is that allspark kernel only supports bits=8, group_size=-1, and desc_act=False. [Misc][Kernel]: Add GPTQAllSpark Quantization vllm-project/vllm#12931

• Exllama vLLM: Very different, structually from the existing v1/v2. We need to benchmark and validate accuracy . If this kernel is good, let's retire Exllama v1/v2.

HF Transformers uses GPTQModel kernels for GPTQ models so this will benefit all HF transformer api/loading.