MELITE v0.2.2. Scale SVC with sklearn Pipeline and refine search grid

[NanoBiostructuresRG]

Summary

This PR prepares MELITE v0.2.2 with an SVC-specific preprocessing update and a focused SVC search-grid refinement.

SVC is now trained and exported as a scikit-learn pipeline:

StandardScaler -> SVC

Scaling is applied only to SVC. RandomForestClassifier and XGBClassifier remain unscaled.

Changes

• Wrapped SVC training in an sklearn Pipeline using StandardScaler followed by SVC.
• Updated SVC hyperparameter grids to use pipeline-compatible parameter names such as svc__C, svc__kernel, svc__gamma, svc__degree, and svc__coef0.
• Updated SVC export so exported SVC models preserve the same StandardScaler -> SVC pipeline used during benchmarking.
• Preserved legacy export compatibility for older SVC parameter dictionaries using unprefixed keys such as C and kernel.
• Kept RandomForestClassifier and XGBClassifier unscaled.
• Added a linear SVC block to the full SVC search grid.
• Adjusted polynomial SVC coef0 grid from 0.02 to 0.2.
• Updated tests covering config, model training, export, and prediction behavior.
• Updated version, changelog, README, citation metadata, and docs references for v0.2.2.

Validation

• python -m pytest tests -p no:cacheprovider --basetemp <temp-dir-outside-repo> -q
  -- 118 passed, 1 warning
• python -c "import melite; print(melite.__version__, melite.__all__)"
  -- 0.2.2 ['Config', 'load_datasets', 'plot_cv_distributions', 'predict', 'version']
• git diff --check
  -- Passed

Notes

• Public API unchanged.
• CLI behavior unchanged.
• Result schema unchanged.
• Prediction remains compatible with exported sklearn pipelines.
• RF and XGB grids were not changed.