Parallelize 5-fold CV and disable QSVC/VQC baselines in quantum classification example

[Copilot]

The 5-fold cross-validation in examples/quantum/classification.py ran folds sequentially and unconditionally trained QSVC and VQC baselines. This PR runs all folds concurrently and comments out the QSVC/VQC paths.

Parallel folds

• Extracted the per-fold body (checkpoint load, classical D=10000, classical D=32, quantum HDC, checkpoint save) into a module-level run_fold(...) so it can be pickled.
• Replaced the for ... in kf.split(...) loop with a ProcessPoolExecutor(max_workers=N_SPLITS) that submits all 5 folds at once and collects results via as_completed.
• Aggregation into the classical_* / quantum_* lists happens after the pool drains, sorted by fold number, so downstream summaries are byte-identical to the sequential run.
• Processes (not threads) because the work is CPU/simulator-bound; expect ~5× memory.

with ProcessPoolExecutor(max_workers=N_SPLITS) as executor:
    futures = [
        executor.submit(run_fold, fold_num, train_idx, test_idx,
                        X_cv_pool, y_cv_pool, X_cv_pool_np, y_cv_pool_np,
                        dimensionality, N_SPLITS)
        for fold_num, (train_idx, test_idx) in fold_splits
    ]
    for future in as_completed(futures):
        fold_num, fold_data = future.result()
        results_by_fold[fold_num] = fold_data

Disabled QSVC/VQC

• Commented out the qiskit imports used only by these baselines (COBYLA, Z/ZZFeatureMap, RealAmplitudes, VQC, QSVC, ComputeUncompute, FidelityQuantumKernel, StatevectorSampler).
• Commented out the QSVC and VQC blocks in run_fold, the corresponding entries in the fold_data checkpoint dict, and — in __main__ — the vqc_* / qsvc_* list inits, aggregation, print_cv_summary calls, ROC data prints (section 6), and ROC plot-point calculations (section 7).
• Classical (D=10000, D=32) and Quantum HDC paths are untouched.