Use raw URLs for split manifest FileObjects

Metadata-only update: use raw Hugging Face URLs for scenes_train.csv and scenes_test.csv in Croissant metadata so checker record generation does not depend on stale git clone caches. Dataset files and benchmark contents are unchanged.

ih-depth.croissant.json

@@ -98,10 +98,7 @@
       "@id": "scenes-train-csv",
       "name": "scenes_train.csv",
       "description": "Scene-level manifest for the 41 training scenes in the frozen benchmark split.",
-      "containedIn": {
-        "@id": "hf-repo"
-      },
-      "contentUrl": "scenes_train.csv",
+      "contentUrl": "https://huggingface.co/datasets/SemilleroCV/ih-depth/raw/main/scenes_train.csv",
       "encodingFormat": "text/csv",
       "sha256": "1472d6695d5dba81b742bd18653e2f8acc74327a0156a96bc2d68c31d4903ca2"
     },
@@ -110,10 +107,7 @@
       "@id": "scenes-test-csv",
       "name": "scenes_test.csv",
       "description": "Scene-level manifest for the 10 test scenes in the frozen benchmark split.",
-      "containedIn": {
-        "@id": "hf-repo"
-      },
-      "contentUrl": "scenes_test.csv",
+      "contentUrl": "https://huggingface.co/datasets/SemilleroCV/ih-depth/raw/main/scenes_test.csv",
       "encodingFormat": "text/csv",
       "sha256": "5575854fbb434343ccec9224e5fe7c19030859e12439190627f7b87f238f5509"
     },