Replaces PR #19. Same 346 newly simulated NCap cap-matrix records, but with `design.design_tool` normalized from 'Qiskit Metal' to 'qiskit-metal' for *all* 464 Sadman-uploaded records (118 already in main + 346 new), so the dataset now reports a single design_tool value across all 894 entries.

Validation summary (vs main):
- Schema OK; sim_options byte-identical to existing.
- 0 negative / non-finite caps, 0 Cauchy-Schwarz violations.
- Monotonicity: 0 / 363 cap_gap->top_to_bottom violations; 0 / 1803 finger_count violations; 27 / 541 finger_length->top_to_bottom violations all at finger_count=1, max 1.99% (~25 mfF, below Ansys LOM noise floor).
- Noise-floor sigma test (robust sigma from LOO on existing 548): median |z| < 1, p90 |z| < 3 on every capacitance.
- 0 duplicates / near-duplicates in the dataset.