[WIP] Evals for HallusionBench

[chiral-carbon]

Adds:

• download instructions for dataset in docs/Evaluation.md
• script to convert HallusionBench.json to llava format for inference
• custom model_vqa_hallusionbench.py script for inference and generating results
• script to convert result file back to .json format to evaluate with official evaluation scripts from HallusionBench GitHub repo

The pipeline works correctly, but the custom vqa script has issues.

---

TODO:

• Investigate and fix model_vqa_hallusionbench.py as it fails to generate response correctly, hence defaulting to model prediction category "2" for all inputs and accuracy = 0