Add support for vLLM KV-cache quantization

[eldarkurtic]

This PR enables using KV-cache quantization with vLLM backend by allowing users to trigger it with the expected vLLM args: kv_cache_dtype and calculate_kv_scales. More details about these two is available at https://docs.vllm.ai/en/stable/serving/engine_args.html#cacheconfig

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.