Expand to new categories in arxiv, test/enhance influence prediction

[benevolentbandwidth-nurtekin]

Currently, Research Agent primarily operates within a limited set of arXiv categories.

As we scale, we want to expand to additional arXiv subject areas and ensure that the impact prediction model remains robust across different research domains.

Goal:

1. Add support for additional arXiv categories (e.g., statistics, mathematics, physics, economics, etc.)
2. Evaluate how influence prediction behaves across these fields
3. Enhance prediction logic if needed to maintain stability and fairness

Key Considerations

1. Citation dynamics vary significantly across domains
2. Conference/journal importance differs by field
3. We should avoid overfitting to CS-style publication patterns
4. Maintain prediction consistency and avoid increased false positives

This issue focuses on expanding coverage while preserving prediction quality.