[Benchmark] Add support for CharXiv benchmark

[MaoSong2022]

OpenAI GPT-4.1 uses CharXiv benchmark to show its vision capability. I belive CharXiv benchmark should be supported by VLMEvalKit.

I am working on supporting this benchmark, any suggestions are welcomed :)

• [x] descriptive_val.tsv file
• [x] reasoning_val.tsv file
• [ ] test