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/index.ts

@@ -1,27 +1,13 @@
-import { scenario as simpleQuery } from './simple-query';
-import { scenario as searchEmbeddings } from './search-embeddings';
-import { scenario as createNode } from './create-node';
-import { scenario as hardModeQuery } from './hard-mode-query';
-import { scenario as updateNode } from './update-node';
-import { scenario as createEdge } from './create-edge';
-import { scenario as queryDimensions } from './query-dimensions';
-import { scenario as getDimension } from './get-dimension';
-import { scenario as dimensionLifecycle } from './dimension-lifecycle';
-import { scenario as youtubeExtract } from './youtube-extract';
-import { scenario as websiteExtract } from './website-extract';
-import { scenario as paperExtract } from './paper-extract';
+import { scenario as focusedGraphWrite } from './focused-graph-write';
+import { scenario as skillGuidedWrite } from './skill-guided-write';
+import { scenario as nodeIndexSearch } from './node-index-search';
+import { scenario as chunkQuoteInsight } from './chunk-quote-insight';
+import { scenario as hubTraversal } from './hub-traversal';
 
 export const scenarios = [
-  simpleQuery,
-  searchEmbeddings,
-  createNode,
-  updateNode,
-  createEdge,
-  queryDimensions,
-  getDimension,
-  dimensionLifecycle,
-  hardModeQuery,
-  youtubeExtract,
-  websiteExtract,
-  paperExtract,
+  focusedGraphWrite,
+  skillGuidedWrite,
+  nodeIndexSearch,
+  chunkQuoteInsight,
+  hubTraversal,
 ];