[Feature] Optimize Tokenzation incluing multi-mode problems, Parser and Formula optimization

[KenelmQLH]

Description

(A clear and concise description of what the feature is.)

• Handle multi-mode problems
  • AST Graph
  • Image
• Handle noise problems when identify $...$ in Parser （need better rules）
• Handle Formula ast problems when identify $AB=BC$ and $123$ （consider preprocessing）

References

• https://huggingface.co/docs/transformers/tasks/image_classification
• http://home.ustc.edu.cn/~huangzhy/files/papers/ZhenyaHuang-SIGIR2020s.pdf