[Perf #478] KernelSVC fit: match libsvm training efficiency without changing semantics

Parent: #478

Canonical v2: Iris Flow fit 1.075 ms vs sklearn 0.904 ms; Digits Flow fit 111.20 ms vs sklearn 50.47 ms. Prediction on Digits is already faster than sklearn, so the dominant deficit is training.

Profile kernel-matrix construction, cache reuse, SMO/working-set selection, shrinking, convergence checks, support-vector compaction, allocation/copy traffic and repeated kernel evaluations. Compare against sklearn/libsvm's algorithmic work, not only wall clock.

Acceptance: stage-level timing and kernel-evaluation counts; explicit cache-hit/miss evidence; scaled sample/feature sweeps; no score/parity regression; fit >=0.95x sklearn on both canonical datasets, target >=1.05x; keep prediction performance at least current.

Start at the KernelSVC fit entry point and profile kernel-matrix construction, cache reuse, SMO/working-set selection, shrinking, convergence checks, support-vector compaction, and allocation or copy traffic. Compare algorithmic work with sklearn/libsvm on the Iris and Digits datasets, recording stage timings, kernel counts, and cache hits or misses. Done means the stated fit-speed, parity, sweep, and prediction-performance acceptance criteria are met.