feat(agents): add Generator-Critic inference demo

[FluffyAIcode]

Summary

• add a real Generator/Critic adversarial inference loop with separate logical histories and two revision rounds
• warm each exact agent context through allens prefill, then run actual generation from the promoted Primary hot KV
• print real model proposal/critique text in terminal while persisting only output hash/length and redacted metrics
• report per-agent TTFT, prefill throughput, decode tok/s, generation ms/token, E2E tok/s, hit source and cache deltas
• report both inference-only KV token hit rate and honest whole-workload hit rate including warmup cost
• extend the Benchmarks UI with KV hit rate, generation latency, Agent/round labels and live latest-stage metrics

Tests

• Agent gate/privacy and report aggregation tests — 5 passed
• full platform-neutral Python suite — 848 passed, 1 skipped
• network/dashboard regression — 10 passed
• Python compile, shell syntax and diff checks pass

Live validation after merge

Run bash scripts/run_agent_gan_demo.sh --rounds 2 --output-tokens 64, inspect real terminal output, then verify the agent_gan_inference_demo history/detail on kakeya.ai.

Made with Cursor

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.