Quiet per-epoch logs in quantum classification example; add fold progress bar and aggregated epochs.json

[Copilot]

examples/quantum/classification.py runs 5 CV folds in parallel and each worker's per-epoch prints interleave chaotically. This PR silences those prints, captures the epoch curves as structured data, and surfaces overall progress via a single progress bar.

Changes

• QuantumClassificationModel.retrain (hdlib/model/classification.py)

  • New optional verbose: bool = True kwarg — default preserves existing behavior.
  • Records per-epoch error on self.retrain_history_ as [{"epoch": int, "error": float}, ...], including epoch 0 and the early-exit path. Callers no longer need to parse stdout.

• examples/quantum/classification.py

  • run_fold calls retrain(..., verbose=False) and adds q_epochs, q_final_epoch, q_final_error to each fold's results dict (also persisted in the existing fold_<n>_results.json).
  • Replaced the --- COMPLETED FOLD n/N --- prints with an in-place stderr progress bar driven by as_completed:

    Progress: [################------------------------]  40.00%  (2/5 folds)
    

  • After all folds finish, writes a single aggregated epochs.json keyed by fold_<n> with each fold's epoch curve and final stats.

Resulting output layout

Unchanged except for the added epochs file:

• fold_1_results.json … fold_5_results.json (now also include q_epochs)
• epochs.json — per-fold epoch history in one place