compatible with openai/tgi/vllm request format

[lkk12014402]

Description

1. compatible with openai/tgi/vllm completion request format/parameters
2. priority default values (openai > tgi > vllm). With same default value, the opea service latency is almost same with native server tgi/vllm
3. replace from langchain_community.llms import VLLMOpenAI with from openai import OpenAI, which can reduce initialization time (0.15s -> 0.01s) and only init once
4. simplify post-process for openai format response (especially remove encode)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
comps/cores/proto/docarray.py	100.00% <100.00%> (ø)

[lkk12014402]

Opea/OpenAI/VLLM/TGI completion request format comparision

OpenAI API Reference - OpenAI API	vLLM vllm/entrypoints/openai/protocol.py	TGI GenerateParameters	OPEA (now)	OPEA (this pr)
model: Union[str	model: str			model: Optional[str] = None # for openai, not used by tgi
prompt: Union[str	prompt: Union[List[int], List[List[int]], str, List[str]]			query: str # alias 'prompt'
best_of: Optional[int] Defaults to 1	best_of: Optional[int] = None	default = "null"		best_of: Optional[int] = 1
echo: Optional[bool] Defaults to false	echo: Optional[bool] = False			echo: Optional[bool] = False
frequency_penalty: Optional[float] Defaults to 0	frequency_penalty: Optional[float] = 0.0	default = "null"	frequency_penalty: Optional[float] = 0.0	frequency_penalty: Optional[float] = 0.0
logit_bias: Optional[Dict[str, int]] Defaults to null	logit_bias: Optional[Dict[str, float]] = None			logit_bias: Optional[Dict[str, float]] = None
logprobs: Optional[int] Defaults to null	logprobs: Optional[int] = None			logprobs: Optional[int] = None
max_tokens: Optional[int] Defaults to 16	max_tokens: Optional[int] = 16	default = "100"	max_tokens: Optional[int] = 1024	max_new_tokens: Optional[int] = 16  # alias 'max_tokens'
n: Optional[int] Defaults to 1	n: int = 1			n: Optional[int] = 1
presence_penalty: Optional[float] Defaults to 0	presence_penalty: Optional[float] = 0.0			presence_penalty: Optional[float] = 0.0
seed: Optional[int] Defaults to null	seed: Optional[int] = Field(None,	default = "null"		seed: Optional[int] = None
stop: Union[Optional[str] Defaults to null	stop: Optional[Union[str, List[str]]] = Field(default_factory=list)	stop: Vec, default = []		stop: Union[Optional[str], List[str], None] = None
stream: Optional[Literal[False]] Defaults to false	stream: Optional[bool] = False		stream: Optional[bool] = False	streaming: Optional[bool] = False  # alias 'stream'
stream_options: Optional[ Defaults to null	stream_options: Optional[StreamOptions] = None			stream_options: Optional[StreamOptions] = None
suffix: Optional[str] Defaults to null	suffix: Optional[str] = None			suffix: Optional[str] = None
temperature: Optional[float] Defaults to 1	temperature: Optional[float] = 1.0	default = "null",	temperature: Optional[float] = 0.01	temperature: Optional[float] = 1.0
top_p: Optional[float] Defaults to 1	top_p: Optional[float] = 1.0	default = "null",	top_p: Optional[float] = 0.95	top_p: Optional[float] = 1.0
user: str Defaults to null	user: Optional[str] = None			user: Optional[str] = None
	more parameters for vllm…..
	…	default = "null"	top_k: Optional[int] = 10	top_k: Optional[int] = None
	…	default = "null"	typical_p: float = 0.95	typical_p: Optional[float] = None
	…	default = "null"	repetition_penalty: Optional[float] = 1.03	repetition_penalty: Optional[float] = None
		do_sample: bool, default = "false"

[lkk12014402]

implement it in another pr