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/chunk-quote-insight.ts

@@ -0,0 +1,19 @@
+import { Scenario } from '../types';
+
+export const scenario: Scenario = {
+  id: 'chunk-quote-insight',
+  name: 'Chunk quote to insight',
+  description: 'Search a focused transcript for a specific quote, then create a grounded insight node from it.',
+  tools: ['searchContentEmbeddings', 'createNode', 'createEdge'],
+  input: {
+    message: 'Search inside this focused transcript for a quote about verification being harder than generating solutions. Quote it briefly, then create a new insight node titled "Lange on verification difficulty" and connect it back to this transcript with explanation "Insight extracted from quoted passage."',
+    focusedNodeQuery: { titleContains: 'When AI Discovers the Next Transformer' },
+  },
+  expect: {
+    toolsCalledSoft: ['searchContentEmbeddings', 'createNode', 'createEdge'],
+    responseContainsSoft: ['Lange on verification difficulty'],
+    maxLatencyMs: 35000,
+    maxTotalTokens: 14000,
+    maxEstimatedCostUsd: 0.14,
+  },
+};