example: add surrogate modeling demonstration with scikit-learn

[Nithurshen]

Summary

This PR adds a new example, surrogate_modeling, which demonstrates how to integrate Mesa with Scikit-learn to create an emulator (surrogate model). This allows users to approximate the behavior of a complex Agent-Based Model (ABM) across a high-dimensional parameter space using only a small number of actual simulation runs.

Motive

Computational expense is a common bottleneck in ABM research, especially during calibration or sensitivity analysis. This PR offloads these concerns to Scikit-learn and Scipy to provide users with a robust workflow for efficient experimentation.

Implementation

• The example is self-contained in examples/surrogate_modeling/ to demonstrate integration without modifying the core library.
• Avoided batch_run and DataCollector in favor of manual Python loops to align with Mesa 4.0 roadmap.

Usage Examples

Users can run the entire pipeline from sampling to prediction with a single command:

python -m examples.surrogate_modeling.analysis

Running simulations for training data...
Surrogate model trained.

Prediction for {'n': 65, 'width': 22, 'height': 22}:
Approximated Gini: 0.6262

Additional Notes

• Added a new optional dependency group surrogate in pyproject.toml including scikit-learn.