Adding RuLES task

[yawen-d]

Evaluation script for LLM RuLES task

• paper: https://arxiv.org/pdf/2311.04235
• github: https://github.com/normster/llm_rules?tab=readme-ov-file

Notes:

• In rules.py, four methods _single_input, _single_eval_message, _single_eval_postprocess, run_pipeline were overwritten.
• Most entries in the dataset are multi-round dialogues that only contain user messages - where the assistant messages were generated on the fly. This is the reason why tasks/RuLES.py#L100 takes in the LLM client to generate the intermediate assistant messages.
• I haven't fully tested the script, but only tested it under debug mode.