Add adversarial verification

[sfloess]

[chatgpt-roadmap]

Future capability

Add an independent verification stage that challenges the proposed consensus and identifies contradictions, unsupported claims, or important omissions.

Potential flow:

Workers -> Arbiter -> Verifier -> Final result

Potential directions:

• Independent verifier model or verifier fleet
• Contradiction detection
• Evidence/claim validation
• Explicit verifier findings in the final response

Not required for PR #2. This is a post-MVP capability to revisit after the current MCP fleet implementation is stable and tested.