SGD and variance reduced SGD for IFs

To complete all methods in Fisher's paper

The issue references implementing SGD and variance reduced SGD for Influence Functions (IFs) to complete methods from a paper. Start by reading the referenced paper to understand the algorithms. Look at existing influence function implementations in the codebase, likely in modules related to influence or valuation. Identify where gradient-based optimization methods are currently used or missing. 'Done' means the new SGD variants are implemented, tested, and integrated with the existing influence function computation pipeline.

由索引模型根据 Issue 内容生成。