Add native MVL-SIB sentence-to-image artifact

README.md

@@ -3299,4 +3299,5 @@ sentence-to-image output is converted to MTEB's retrieval schema. Query
 texts, candidate sets, their order, and labels are unchanged. Images are
 downsampled with the paper's aspect-preserving 640x480 preprocessing.
 Only IDs and packaging are added: candidate images are stored as
-`top_ranked`, and all languages reuse one corpus config.
+`top_ranked`, and every language-specific corpus config references
+the same physical file.