test(core): tolerate grader timing jitter

[christso]

Summary

• Relax the flaky per-grader timing assertions to require positive recorded duration instead of an exact lower bound matching setTimeout delays.
• Keep the existing timestamp assertions that prove duration_ms is internally consistent with started_at/ended_at timing.

Root cause

Main CI failed in GitHub Actions run https://github.com/EntityProcess/agentv/actions/runs/27316820557/job/80699130076 because packages/core/test/evaluation/orchestrator.test.ts asserted durationMs >= 50 after awaiting a 50ms timer. CI recorded 49ms. The orchestrator records millisecond wall-clock timestamps before and after the evaluator call, so timer/clock granularity can make the measured elapsed time one millisecond below the requested delay even though timing metadata is present and correct.

Verification

• cd packages/core && bun test test/evaluation/orchestrator.test.ts -t "includes per-grader timing in scores" repeated 30x
• cd packages/core && bun test test/evaluation/orchestrator.test.ts -t "includes per-grader timing even when evaluator fails" repeated 30x
• cd packages/core && bun test test/evaluation/orchestrator.test.ts
• bun --filter @agentv/core test
• Pre-push hook: typecheck + biome check .

[cloudflare-workers-and-pages]

Deploying agentv with    Cloudflare Pages

Latest commit:	60e4eb6
Status:	✅  Deploy successful!
Preview URL:	https://a04bb565.agentv.pages.dev
Branch Preview URL:	https://fix-orchestrator-timing-flak.agentv.pages.dev

View logs