Add SVR baseline model

[abdelghanibelgaid]

Summary

This PR adds a Support Vector Regression (SVR) baseline to CY-Bench, following the same structure as the existing scikit-learn baselines.

Changes

• Added SVRModel in cybench/models/svr_model.py

  • Uses sklearn.svm.SVR with an RBF kernel (C=1.0, epsilon=0.1) as the base estimator.
  • Reuses the BaseSklearnModel feature pipeline (same design as SklearnRidge and SklearnRF).
  • Enables the internal hyperparameter search with a small parameter grid:
    • estimator__C: [0.1, 1.0, 10.0]
    • estimator__epsilon: [0.01, 0.1, 0.5]

• Added residual variant:

  • SVRRes in cybench/models/residual_models.py, wrapping SVRModel inside the existing ResidualModel logic.

• Registered the new models in the benchmark runner:

  • Imported SVRModel and SVRRes in cybench/runs/run_benchmark.py.
  • Exposed them in _BASELINE_MODEL_CONSTRUCTORS as "SVRModel" and "SVRRes" so they can be run as standard baselines.

• Updated plotting utilities:

  • Extended model_short_names in cybench/runs/results_plots.py with:
    • "SVRModel": "SVR"
    • "SVRRes": "SVR-Res"

Notes

• The implementation mirrors the pattern used for SklearnRandomForest and XGBoostModel, so SVR shares the same feature engineering and evaluation pipeline.
• No additional dependencies are introduced; SVR is part of the existing scikit-learn dependency.

Checklist

• Added SVRModel with BaseSklearnModel feature pipeline
• Added SVRRes residual wrapper
• Registered models in run_benchmark.py
• Added short names in results_plots.py
• Updated model docs if you prefer to list SVR explicitly

Closes #378.