Request page images explicitly from docling-serve

CHANGELOG.md

@@ -12,6 +12,10 @@
 
 - `provider: vllm` is text-only unless `embeddings.model.multimodal: true` is set. Existing multimodal vLLM configs must add the flag; `multimodal` is not part of the stored embedding identity, so changing it raises no drift error — re-ingest or `rebuild` to add or drop picture chunks.
 
+### Fixed
+
+- `docling-serve` converter requests `include_page_images`, so `generate_page_images` produces page rasters on docling-serve versions that gate them behind that flag.
+
 ## [0.60.0] - 2026-06-22
 
 ### Added

haiku_rag_slim/haiku/rag/converters/docling_serve.py

@@ -113,6 +113,7 @@ class DoclingServeConverter(DocumentConverter):
             "images_scale": str(opts.images_scale),
             "image_export_mode": "referenced",
             "include_images": str(pictures != "none").lower(),
+            "include_page_images": str(opts.generate_page_images).lower(),
             "do_picture_description": str(runs_vlm).lower(),
             "target_type": "zip",
         }
@@ -187,10 +188,9 @@ class DoclingServeConverter(DocumentConverter):
             for picture in doc_json.get("pictures") or []:
                 _inline(picture.get("image"))
 
-            # Docling-serve always bundles page rasters when
-            # image_export_mode="referenced". Honor the local
-            # ``generate_page_images`` flag by stripping them when the user
-            # didn't ask for whole-page images.
+            # Page rasters are bundled only when ``include_page_images`` is set
+            # on the request, which mirrors ``generate_page_images``. Strip any
+            # that slip through when the user didn't ask for whole-page images.
             keep_page_images = (
                 self.config.processing.conversion_options.generate_page_images
             )