refactor to move under matrix.client.query_llm

[dongwang218]

Why ?

query_llm was in the matrix.app_server.llm, now move it to client, so that we can add other clients.

How ?

similarly move EndpointCache to client.
move the llm default parameters into a separate file llm_config.py

Test plan

matrix deploy_applications --applications "[{'model_name': '/datasets/pretrained-llms/Llama-3.1-8B-Instruct', 'use_grpc': 'true', 'min_replica': 8, 'model_size': '8B', 'name': '8B_grpc'}]"
matrix check_health --app_name 8B_grpc