Croissant: add rai:hasSyntheticData, prov:wasGeneratedBy; rename dataLimitation -> dataLimitations

Croissant: semantic_seg = per-pixel category IDs (not instance IDs); keyed to per-scene COCO categories

Croissant: fix semantic_seg shape (12,480,640) uint16, drop spurious channel dim

Fix Croissant: ISO-8601 dates, isArray/arrayShape for array fields

Update README+Croissant for 167 HSSD scenes / all-semantics / 3.3 TB; drop 3 non-HSSD test scenes (apartment_1, skokloster-castle, van-gogh-room) for license incompatibility

Add dataset card and Croissant metadata with RAI fields for NeurIPS 2026 D&B submission