feat: Add AGI Architecture Research Packet (Bounty Submission)

[parastejpal987-cmyk]

Addresses Bounty Issue #5 (Collect and compare AGI architecture proposals).

This pull request introduces the complete research packet into research/ai_generated_agi_architectures/.

Deliverables Included:

1. README.md: Executive summary and methodology.
2. prompts.md: The exact system prompts used to elicit the designs.
3. raw_outputs/: 8 distinct architecture proposals generated by the following models:
   • GPT-4o
   • Claude 3.5 Sonnet
   • Gemini 1.5 Pro
   • Llama 3 70B
   • Mistral Large
   • DeepSeek-V2
   • Command-R+
   • Grok-1.5
4. comparison.csv: Matrix evaluating the architectures across Reasoning, Memory, Modularity, and Alignment.
5. summary.md: Synthesis of common patterns, specifically regarding the decline of pure next-token prediction and the rise of decentralized hypervisor-based safety.

All files strictly adhere to the requested format and folder structure. Please review for bounty acceptance.

[parastejpal987-cmyk]

I've resolved the failing CI test! The test_product_cli_routes_codex_auth_provider test was failing because runner=subprocess.run was early-bound as a default argument, meaning monkeypatch.setattr in the tests couldn't actually intercept the subprocess call. The test was trying to execute the real codex binary on the CI runner and failing. I've switched it to use late-binding (runner=None), which allows the test mocks to work correctly. The test suite is now completely green!