fix: coerce stringified arrays in onboarding overview output (#32)

Opening an onboarding theme could fail with "LLM output failed schema
validation": the fast-tier model occasionally emits the key_points and
topics_covered tool fields as JSON-encoded strings instead of real
arrays. Same failure class the quiz schema already guards against with
z.preprocess.

- onboardingOverviewSchema: parse "[...]" strings back to arrays before
  validation; split a bullet/newline string of key points as a fallback;
  keep the first 5 key points on overage instead of failing the user.
  Genuinely bad output (too few points, non-JSON strings, wrong types)
  still fails validation.
- emit_onboarding_overview tool: descriptions now state the fields must
  be JSON arrays, never strings (prompt-side nudge).
- 5 regression tests reproducing the exact #32 payload shapes
  (npm test 139/139).

Closes #32

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/lib/__tests__/llmSchemas.test.js

@@ -9,6 +9,7 @@ import {
   customTopicSchema,
   graphActionsSchema,
   proposeGraphDeltaSchema,
+  onboardingOverviewSchema,
 } from '../llmSchemas';
 
 const sampleTopic = {
@@ -192,3 +193,56 @@ describe('proposeGraphDeltaSchema', () => {
     ).toThrow();
   });
 });
+
+describe('onboardingOverviewSchema (issue #32 — fast-tier stringified arrays)', () => {
+  const base = {
+    title: 'Privacy',
+    what_it_is: 'How we handle data.',
+    why_it_matters: 'You touch personal data weekly.',
+  };
+  const points = ['Point one', 'Point two', 'Point three'];
+  const topics = [{ topic_id: 'avg', label: 'AVG' }];
+
+  it('accepts well-formed output', () => {
+    const r = onboardingOverviewSchema.parse({ ...base, key_points: points, topics_covered: topics });
+    expect(r.key_points).toEqual(points);
+    expect(r.topics_covered).toEqual(topics);
+  });
+
+  it('coerces JSON-stringified arrays back to arrays (the exact #32 failure)', () => {
+    const r = onboardingOverviewSchema.parse({
+      ...base,
+      key_points: JSON.stringify(points),
+      topics_covered: JSON.stringify(topics),
+    });
+    expect(r.key_points).toEqual(points);
+    expect(r.topics_covered).toEqual(topics);
+  });
+
+  it('splits a bullet/newline string of key points as a fallback', () => {
+    const r = onboardingOverviewSchema.parse({
+      ...base,
+      key_points: '- Point one\n• Point two\n3. Point three',
+      topics_covered: topics,
+    });
+    expect(r.key_points).toEqual(points);
+  });
+
+  it('keeps the first 5 key points instead of failing on overage', () => {
+    const seven = Array.from({ length: 7 }, (_, i) => `P${i + 1}`);
+    const r = onboardingOverviewSchema.parse({ ...base, key_points: seven, topics_covered: topics });
+    expect(r.key_points).toEqual(['P1', 'P2', 'P3', 'P4', 'P5']);
+  });
+
+  it('still rejects genuinely bad output', () => {
+    expect(() =>
+      onboardingOverviewSchema.parse({ ...base, key_points: ['only', 'two'], topics_covered: topics }),
+    ).toThrow();
+    expect(() =>
+      onboardingOverviewSchema.parse({ ...base, key_points: points, topics_covered: 'not json at all' }),
+    ).toThrow();
+    expect(() =>
+      onboardingOverviewSchema.parse({ ...base, key_points: 42, topics_covered: topics }),
+    ).toThrow();
+  });
+});