[ci] benchmark react

[zhulinJulia24]

1. react benchmark test to use 8 gpu in parallel and add benchmark script test into daily test
2. add llama3.1 into evaluation, function and benchmark test https://github.com/zhulinJulia24/lmdeploy/actions/runs/10109254518
3. add timecouter into evaluation https://github.com/zhulinJulia24/lmdeploy/actions/runs/10143171693
4. rename w4a16 into 4bits
5. fix condition: XX is false condition into not XX
6. remove --cap by using chat-template

add stable testflow
add gptq quantization testcase
change test basic docker image to make test step more simple