feat: sync runtime search and schema quality updates from app repo

- port retrieval, validation, and eval improvements relevant to os
- align prompts and dimensions with the flat single-agent model
- replace the old eval suite with the focused core scenarios

Generated with Codex

tests/evals/scenarios/focused-graph-write.ts

@@ -0,0 +1,20 @@
+import { Scenario } from '../types';
+
+export const scenario: Scenario = {
+  id: 'focused-graph-write',
+  name: 'Focused graph write',
+  description: 'Create one new node from the focused transcript and connect it back without unnecessary retrieval.',
+  tools: ['createNode', 'createEdge'],
+  input: {
+    message: 'Create a new node titled "Lange: Verification bottleneck" for the claim that generating many solutions is easier than verifying them, then connect it to this focused transcript with explanation "Claim extracted from this transcript source."',
+    focusedNodeQuery: { titleContains: 'When AI Discovers the Next Transformer' },
+  },
+  expect: {
+    toolsCalledSoft: ['createNode', 'createEdge'],
+    toolsNotCalledSoft: ['readSkill', 'queryNodes', 'searchContentEmbeddings'],
+    responseContainsSoft: ['Lange: Verification bottleneck'],
+    maxLatencyMs: 25000,
+    maxTotalTokens: 9000,
+    maxEstimatedCostUsd: 0.08,
+  },
+};