(Approximate) Maximum Influence Perturbation

As described in _Broderick, Giordano, and Meager, ‘An Automatic Finite-Sample Robustness Metric’._

The issue references a paper on 'An Automatic Finite-Sample Robustness Metric' and mentions 'Maximum Influence Perturbation'. Start by reading the cited paper to understand the algorithm. Look for existing influence function implementations in pyDVL, likely in modules like `influence` or `valuation`. Determine what inputs the perturbation method needs and how it integrates with current data valuation workflows. Check for tests related to influence or robustness to see the expected output format.

由索引模型根据 Issue 内容生成。