Fix average secrets per sample: 1.29 -> 1.49 (matches dataset_stats.json)

Update v1.2: complete multi_secret ground-truth labels

Add Synthetic Sensitive Data in Source Code (N=300)

initial commit