Upload docs/DATA_SCHEMA.md with huggingface_hub

docs/DATA_SCHEMA.md

@@ -0,0 +1,15 @@
+# Data Schema
+
+            Each row in the parquet file contains:
+
+            - `text`: the textual content
+            - `source`: the originating source or collection label
+            - `domain`: the broad category of the row
+            - `script`: script label such as `devanagari`, `latin`, or `mixed`
+            - `lang`: language tag used during preprocessing
+            - `date_collected`: collection or extraction date
+            - `license`: per-row license label
+
+            ## Notes
+            - These fields are intentionally simple so the dataset can be loaded in most NLP pipelines without custom parsing.
+            - Script and language labels are heuristic and are best treated as descriptive metadata rather than ground truth.