[ci] regular update

[zhulinJulia24]

1. [tested] add evaluation function test into daily test, only run 50 cases for each benchmark dataset
2. [tested] react benchmark test, because more than one logs need to attached to benchmark report so move allure report attachment action into utils
3. [tested] fix do_sample param related testcase
4. [tested] react env paramter of env.COV_PARAM in daily test
5. [tested] change installation method of autoawq_kenerls
6. [tested] add more test models into benchmark test
7. [Not fully verified] try to use pytest_cache when rerun pipelines

[lvhan028]

It is possible that the reponse.text is different for batch inference, because they are performed in continuous batching rather than fix batch size
Let's change it to the following:

response = [pipe('Shanghai is', gen_config=gen_config) for i in range(3)]
result = response[0].text == response[1].text and response[1].text == response[2].text