Integrate TensorRT-LLM

[yunfeng-scale]

Integrating TensorRT-LLM with LLM engine. there are lots of caveats in this integration:

• Stop word not working for TRT-LLM Triton backend Param "stop_words" not respected in v2/models/ensemble/generate endpoint triton-inference-server/tensorrtllm_backend#57
• Log prob not returned Return log probabilities for tokens NVIDIA/TensorRT-LLM#238
• Need to verify definitions of presence_penalty and repetition_penalty that TRT-LLM supports
• Need to send back only newly generated tokens How to get the newly generated tokens only? NVIDIA/TensorRT-LLM#227
• Can't tell end of stream Flag indicate end of stream NVIDIA/TensorRT-LLM#240

fixes #363

[yixu34]

Some minor comments.

Can we create a github issue for visibility and link it to this PR?