fix(graph): filter dangling edges before passing to cytoscape-dagre

Cytoscape-dagre crashes with "Cannot set properties of undefined
(setting 'order')" when an edge's source or target is not present in
the cytoscape node set.  Both backends can produce such edges in edge
cases (ON_PAGE edges referencing a page_no absent from the pages list,
DERIVED_FROM edges crossing document boundaries in Neo4j).  When
cytoscape-dagre adds these edges to its internal dagre/graphlib graph,
graphlib auto-creates a skeleton node entry with undefined data; dagre's
ordering phase then tries to set `.order` on that undefined object and
throws.

The fix builds a nodeIds Set from the returned nodes before constructing
the elements array, then filters every edge so only those with a valid
source AND target are forwarded to cytoscape.  Dangling edges have no
visual meaning anyway, so this is purely defensive.

https://claude.ai/code/session_01PjZeeTBqERYDPs59RhXqCz

.github/workflows/docling-compat.yml

@@ -27,7 +27,7 @@ jobs:
         with:
           python-version: "3.12"
           cache: pip
-          cache-dependency-path: document-parser/requirements-test.txt
+          cache-dependency-path: document-parser/requirements.txt
 
       - name: Install system dependencies
         run: sudo apt-get update && sudo apt-get install -y --no-install-recommends poppler-utils
@@ -35,8 +35,8 @@ jobs:
       - name: Install pinned dependencies
         run: |
           pip install --upgrade pip
-          pip install -r requirements-test.txt
+          pip install -r requirements.txt
-          pip install httpx
+          pip install pytest pytest-asyncio httpx
 
       - name: Upgrade docling to latest
         id: versions

frontend/src/features/analysis/ui/GraphView.vue

@@ -238,6 +238,14 @@ async function renderGraph(): Promise<void> {
   )
   parentMap.value = computedParentMap
 
+  // Guard against dangling edges (source/target not in the node set). Both
+  // backends can emit them in edge cases — e.g. ON_PAGE edges whose page_no
+  // doesn't match any returned Page node, or DERIVED_FROM edges crossing
+  // document boundaries in Neo4j. Cytoscape-dagre auto-creates a skeleton
+  // graphlib entry for the missing endpoint with undefined data, then crashes
+  // in its ordering phase when it tries to set `.order` on that entry.
+  const nodeIds = new Set(payload.value.nodes.map((n) => n.id))
+
   const elements = [
     ...payload.value.nodes.map((n) => ({
       data: {
@@ -256,14 +264,16 @@ async function renderGraph(): Promise<void> {
         raw: n,
       },
     })),
-    ...payload.value.edges.map((e) => ({
+    ...payload.value.edges
-      data: {
+      .filter((e) => nodeIds.has(e.source) && nodeIds.has(e.target))
-        id: e.id,
+      .map((e) => ({
-        source: e.source,
+        data: {
-        target: e.target,
+          id: e.id,
-        type: e.type,
+          source: e.source,
-      },
+          target: e.target,
-    })),
+          type: e.type,
+        },
+      })),
   ]
 
   cy.value = cytoscape({