Initial enablement for text-embedding

[libinta]

This is the initial PR for text-embedding with --dtype bfloat16. The following models are tested with 1x and 2x mpi/ray

model
BAAI/bge-base-en-v1.5
BAAI/bge-multilingual-gemma2
intfloat/e5-mistral-7b-instruct
ssmits/Qwen2-7B-Instruct-embed-base
Alibaba-NLP/gte-Qwen2-7B-instruct

There is no Lora support yet.
has dependency with HabanaAI/vllm-hpu-extension#87

offline test use script from https://docs.vllm.ai/en/latest/getting_started/examples/embedding.html with adding dtype="bfloat16" in LLM call
sample cmds such as
python testscript.py
VLLM_SKIP_WARMUP=true python testscript.py

online test use client script from https://docs.vllm.ai/en/latest/getting_started/examples/openai_embedding_client.html
server sample cmds such as
vllm serve Alibaba-NLP/gte-Qwen2-7B-instruct --task embed --dtype bfloat16
VLLM_SKIP_WARMUP=true VLLM_PROMPT_USE_FUSEDSDPA=true vllm serve Alibaba-NLP/gte-Qwen2-7B-instruct --task embed --dtype bfloat16