feat: implement Story 3.2 — AI diagram generation from natural language

Add complete AI-powered diagram generation pipeline: natural language input
→ type inference → graph patch generation → validated canvas render with
ELK.js layout animation. Includes adversarial code review fixes for
diagramType enum validation, duplicate ID detection, tool part history
preservation, PATCH error handling, and graphData structural validation.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

packages/api/src/modules/diagram/router.ts

@@ -32,12 +32,22 @@ export const updateDiagramBodySchema = z
     title: z.string().min(1).max(255).optional(),
     projectId: z.string().nullable().optional(),
     sortOrder: z.number().int().min(0).optional(),
+    graphData: z
+      .object({
+        meta: z.record(z.string(), z.unknown()).optional(),
+        nodes: z.array(z.record(z.string(), z.unknown())),
+        edges: z.array(z.record(z.string(), z.unknown())),
+        pools: z.array(z.record(z.string(), z.unknown())).optional(),
+        groups: z.array(z.record(z.string(), z.unknown())).optional(),
+      })
+      .optional(),
   })
   .refine(
     (data) =>
       data.title !== undefined ||
       data.projectId !== undefined ||
-      data.sortOrder !== undefined,
+      data.sortOrder !== undefined ||
+      data.graphData !== undefined,
     { message: "At least one field must be provided" },
   );