fix(multimodal): collect all surrounding text as context instead of only the nearest segment

Previously, `get_image_context` only picked the single text item
immediately adjacent to the image (via `break`). When users placed
the image at the end of the `response` list (e.g. after a task
description + source + target), `context_above` would contain only
the last string (often a single word like "Off" or "Custom") while the
task description was silently dropped. This caused `ImageCoherenceGrader`
and `ImageHelpfulnessGrader` to receive near-empty context, resulting in
unstable and severely low scores (1–2) for the same content that scored 5
when the image was placed first.

The fix concatenates all text segments before/after the image,
ensuring task instructions and surrounding content are always fully
passed to the model regardless of image position.

Made-with: Cursor

Author: XiaoBoAI

openjudge/graders/multimodal/_internal/context_utils.py

@@ -58,23 +58,24 @@ def get_image_context(
         ...     max_context_size=500
         ... )
     """
-    context_above = None
-    context_below = None
-
-    # Find context above (last text before image)
-    for i in range(image_index - 1, -1, -1):
-        if isinstance(content_list[i], str):
-            context_above = content_list[i]
-            if max_context_size and len(context_above) > max_context_size:
-                context_above = context_above[-max_context_size:]
-            break
-
-    # Find context below (first text after image)
-    for i in range(image_index + 1, len(content_list)):
-        if isinstance(content_list[i], str):
-            context_below = content_list[i]
-            if max_context_size and len(context_below) > max_context_size:
-                context_below = context_below[:max_context_size]
-            break
+    # Collect all text segments above the image (in order)
+    above_parts = [
+        content_list[i]
+        for i in range(image_index)
+        if isinstance(content_list[i], str)
+    ]
+    context_above = "\n".join(above_parts) if above_parts else None
+    if context_above and max_context_size and len(context_above) > max_context_size:
+        context_above = context_above[-max_context_size:]
+
+    # Collect all text segments below the image (in order)
+    below_parts = [
+        content_list[i]
+        for i in range(image_index + 1, len(content_list))
+        if isinstance(content_list[i], str)
+    ]
+    context_below = "\n".join(below_parts) if below_parts else None
+    if context_below and max_context_size and len(context_below) > max_context_size:
+        context_below = context_below[:max_context_size]
 
     return context_above, context_below