DNNaic canonical simulation dataset (ADZE-based introgression prediction)

Canonical simulation dataset for the DNNaic project (ADZE-based prediction of genetic introgression). This is the large-data companion to the code at github.com/yspennstate/ADZEProjects — it holds the ~3.2 GB of per-replicate simulation outputs that exceed GitHub's file-size limits. Contents (in simulation_data.tar.gz): the 3,200-replicate canonical training CSV (simulation_dataset.csv, ~367 MB); the expanded 9,600-replicate set (simulation_dataset_9600.csv plus simulation_dataset_extra_round1/2.csv) behind the honest headline; the split manifest; and the leakage-free per-replicate arrays consumed by the canonical trainer (regen_full/: X, direction, magnitude, groups, design, columns). Data contract (verified): 3,200 replicates × 198 rows = 633,600 rows; direction classes A/B/C/D = 900/900/900/500; 2,700 positive / 500 control; rarefaction depth g = 2..199. Use: download simulation_data.tar.gz and run DNNaic/scripts/fetch_simulation_data.py fetch from the repository, or extract it into DNNaic/data/data/simulation_data/.