[Perf #478] RandomForest Digits: remove per-tree training overhead and improve scaling

Parent: #478

Canonical v2 Digits: Flow fit 93.14 ms vs sklearn 28.95 ms; Flow predict 0.608 ms vs sklearn 0.919 ms. Prediction already wins; training is the entire regression. Iris end-to-end wins ~9.9x.

Profile bootstrap/sample generation, RNG, per-tree setup, tree fit, feature subsampling, allocation/free, tree storage and opportunities for parallel or batched work. Reuse #482 findings where the tree builder is shared.

Acceptance: tree-by-tree timing and allocation profile; distinguish forest orchestration from tree-core cost; scaled n_estimators/sample/feature sweeps; preserve bootstrap/RNG semantics and learned-state diagnostics; Digits fit >=0.95x first, target >=1.05x, without sacrificing the current prediction win.

Start by profiling RandomForest Digits training across bootstrap/sample generation, RNG, per-tree setup, tree fitting, feature subsampling, allocation, and tree storage. Reuse findings from issue #482 to separate forest orchestration from shared tree-core cost, then run the requested estimator, sample, and feature sweeps. Done means the timing and allocation profile is documented, semantics and diagnostics are preserved, and Digits fit reaches at least 0.95x without losing the prediction win.