Epic: Model selection and tuning — GridSearchCV revival decision, RFE

[Mec-iS]

Tracking issue for model-selection work. First task is a decide-or-delete call; code follows only after that.

Member issues

• Should we implement SVM grid search & GridSearchCV? (preserved dead code from #398 cleanup) #429 — implement SVM grid search & GridSearchCV? (decision + revival)
• Implement Recursive Feature Elimination (RFE) algorithm #61 — Recursive Feature Elimination (RFE)

Step 0: decide-or-delete for the preserved dead code

~80% of grid-search material survives in-tree but is stale:

• src/model_selection/hyper_tuning/grid_search.rs — near-complete GridSearchCV, not compiled (module decl commented out in mod.rs, references old paths and a generic CrossValidationResult<T> that no longer matches).
• src/svm/search/svc_params.rs — fully commented out, references removed APIs.
• src/svm/search/svr_params.rs — compiles but unused outside its own tests.
  Either revive under this epic or delete in a cleanup PR. Leaving it uncompiled invites drift.

Why grouped

RFE (#61) is a loop over cross_validate + estimator traits. Its API depends on what the tuning layer looks like after the #429 decision.

Current state (main @ a955334)

model_selection offers KFold, train_test_split, cross_validate, cross_val_predict. The api::SupervisedEstimator/Predictor traits are working plumbing for both members.

RFC requirements

Post a short RFC comment covering: revive-vs-delete outcome for hyper_tuning, public API of the search layer, parameter-grid representation (no macros), serde story.

Acceptance criteria

• Decide-or-delete done for preserved files
• RFC agreed by a maintainer
• Members implemented or explicitly descoped
• CHANGELOG updated