[CI] update testcases' baseline

[zhulinJulia24]

1. fix baseline because of lmdeploy/vllm/transformers version change
2. add more bench including llmjudger and cascasd judger datasets into daily testcase
3. add longtext datasets (using test range 0:16) into daily test
4. support llm_evaluator use local gpu by changing opencompass/tasks/openicl_eval.py

[tonysy]

LGTM