Added SVRG optimizer documentation and tests for stability and convergence

[Vishal-sys-code]

This PR introduces a new Stochastic Variance Reduced Gradient (SVRG) optimizer to the statsmodels_sgd repository, including comprehensive documentation and unit tests for accuracy and stability.

Key Changes

1. SVRG Optimizer Implementation:

   • Added an SVRG optimizer in optimizers.py, leveraging variance reduction for improved convergence stability and speed compared to standard SGD.
   • Supports configurable learning rate (lr) and inner-loop iterations (n_inner), providing flexibility for various regression and classification tasks.

2. Documentation:

   • Created svrg_optimizer.md in the docs folder.
   • Detailed documentation covering the optimizer’s purpose, configuration options, usage examples, and integration steps.

3. Unit Tests:

   • Added thorough tests in tests/test_optimizers.py to verify parameter updates, stability, and convergence properties of SVRG.
   • Tests are designed to cover various scenarios, ensuring the robustness of the implementation.

This addition enhances statsmodels_sgd by introducing a more efficient optimizer option, making it suitable for large datasets where variance reduction can significantly improve results.

Please review and merge to make this improvement available in the main repository.