Fix data pipeline: recover Q25, Q30, Q37, Q63 + GPT-4 for 13 questions

- Normalize gpt4o → gpt4 across all schemas (recovers GPT-4 for Q34-Q46, Q63)
- Handle contributions schema (recovers Q25 Message to the Shadows)
- Fix MoltBook question mapping (recovers Q30 MoltBook Emergence)
- Handle nested questions array schema (recovers Q37 Reverse Engineer Predictions)
- Handle object-valued responses {response, length, timestamp} (recovers Q63)
- Re-run invariance analysis: 232 responses from 42 questions (was 206/38)
- All metrics improved: KNN purity now 0.77-0.84 (was 0.68-0.72)

61/64 questions with data, 1866 total responses (was 58/64, 1742)
Only Q20, Q21, Q64 remain missing (no probe files exist)

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

Author: Jamie Nucho