Feat: Add accuracy evaluation for LLMs (GPQA, AIME, HLE etc.)

[nvzhihanj]

Add datasets, post-processing scripts, environment (likely docker) to evaluate the accuracy based on the output collected from the inference endpoints.
List of accuracy eval to add:

• GPQA
• MMLU, MMLU-Pro
• AIME
• MATH500
• HLE
• Health Bench
• TBD