Add TMLU Benchmark Dataset

[adamlin120]

This PR adds support for the TMLU ("Measuring Taiwanese Mandarin Language Understanding" by Chen et al) benchmark dataset.

Summary

• Adds a new dataset tmlu with 2,981 multiple-choice questions across 37 subjects
• Uses the TMLU dataset hosted on Hugging Face
• Supports evaluating Taiwanese Mandarin language understanding using log-likelihood multiple choice scoring
• Includes tasks for each TMLU subject, e.g. tmlu_geography, tmlu_physics, etc.
• Enables reproducing results from the Open Taiwan LLM leaderboard

Checklist

• [x] Referenced the original TMLU paper
• [x] Checked the TMLU reference implementation
• [x] Verified the Hugging Face dataset matches the data used in the TMLU paper

Please let me know if you have any other suggestions or feedback on this PR!

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
2 out of 3 committers have signed the CLA.

:white_check_mark: lintangsutawika
:white_check_mark: adamlin120
:x: Yen-Ting Adam, Lin

---

Yen-Ting Adam, Lin seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[lintangsutawika]

@adamlin120 made some adjustments to this PR here https://github.com/adamlin120/lm-evaluation-harness/pull/1

[lintangsutawika]

@adamlin120 just need your help to run pre-commit run --all-files and it should be good!