Adding four tasks from RULER

[rbiswasfc]

This PR adds four tasks from the RULER benchmark. Specifically, these are the tasks:

• QA2 (hotpotqa after adding distracting information)
• Multi-hop Tracing: Variable Tracking (VT)
• Aggregation: Common Words (CWE)
• Multi-keys Needle-in-a-haystack (NIAH)

Currently, each task is having 4k context length - which can be adjusted as needed.

[rbiswasfc]

Running with default setting with Qwen2-1.5B-Instruct model (e.g. python eval.py --tasks rulercwe --checkpoint_path checkpoints/Qwen/Qwen2-1.5B-Instruct/model.pth ) I got these scores:

• rulerqa: StringMatch_score: 38.8
• rulerniah: StringMatch_score: 89.00
• rulervt: StringMatch_score: 77.96
• rulercwe: StringMatch_score: 31.08