[ci] add w8a8 and internvl2.5 models into testcase

[zhulinJulia24]

1. update test env installation, use /root/packages/flash_attn-*.whl instead of paticular version
2. add gemma-2-27-it and qwen2.5-7b-instruction and some w8a8 models into evalation test
3. add w8a8 models into benchmark test
4. add more models into function test, including internvl2.5 and Molmo-7B-D-0924.
5. add some models(tp = 4) into testcase
6. use local picture and vedios instead of https url to make mllm testcase more stable.

[zhulinJulia24]

testing