cover symbolic_trace by tracing several vision models in tests

[jzhang533]

selected 10 vision models to trace, 8 of them succeed.
2 of them failed because of

• failed on DenseNet, since it calls into _C_ops.
• failed on ShuffleNet, since it calls into x.shape[0:4]

[gglin001]

LGTM! 另外对于 test 我们需要 “完全“ 保持 pytest 的写法吗? 还是不限制(unitets or pytest)

[jzhang533]

LGTM! 另外对于 test 我们需要 “完全“ 保持 pytest 的写法吗? 还是不限制(unitets or pytest)

我对这个没有特别的想法。也许未来需要，我在想，现在的阶段，我们可以不做限制，