research: compare 8 AI-generated AGI architecture proposals

[aibitious]

Addresses #5.

This PR adds an auditable research packet under research/ai_generated_agi_architectures/ based on one comparable architecture prompt executed across 8 distinct model/system families from 7 providers.

Included:

• 8 separately preserved raw model outputs
• exact comparable prompt and prompt SHA-256
• structured comparison.csv across all 11 requested dimensions
• cross-model summary and concrete implementation-oriented synthesis
• model/provider attribution, collection dates, API call identifiers, output hashes, token usage and edit disclosure
• deterministic acceptance report (PASS, 8/8 families, no missing required items)
• packet checksum

System families represented: Anthropic Claude Sonnet 5, DeepSeek v4, Gemini 3.5, Gemini 3.6, OpenAI GPT-OSS via Groq, Mistral Ministral, OpenAI GPT-5.6, and xAI Grok 4.5.

The raw outputs were collected through authorized provider API access. No API keys, account tokens, private system prompts, paid-account screenshots, or provider-private material are included. Raw outputs are kept separate from analysis and are not rewritten.

Canonical research run: b2fcca6e-fad0-483a-8f2c-c5f559900965.

The packet's local deterministic acceptance check reports PASS with 8 raw outputs and 8 distinct system families.