fix(agents): bound Critic evidence prefill context

[FluffyAIcode]

Summary

• preserve full Generator output to EOS in Terminal
• build a bounded Critic evidence window from Generator beginning and conclusion, defaulting to 128 content tokens
• include full/evidence/omitted token counts and Generator EOS completion status
• instruct Critic that evidence omission is a transport constraint, not Generator incompleteness
• emit an allens Prefill heartbeat every 30 seconds with elapsed time
• expose evidence metrics in benchmark phase cards and reports

Root cause

A complete Riemann answer produced a 1,166-token Critic prompt. The 16GB allens worker remained active but could not finish Prefill within 900 seconds. Raising timeouts does not make the interaction usable.

Tests

• evidence full/omitted/budget validation
• heartbeat elapsed progress
• stage evidence metrics and Critic gate tests — 11 passed
• full platform-neutral suite — 857 passed, 1 skipped
• Python compile and diff checks pass

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.