feat(benchmarks): add memory benchmark suite

[drewstone]

What changed

• Adds memory benchmark task kinds for ingest, recall, temporal update, forgetting, reasoning, summarization, recommendation, and multi-party attribution.
• Adds INDUSTRY_MEMORY_BENCHMARKS, KnowledgeMemoryBenchmarkCase, buildIndustryMemoryBenchmarkSmokeCases(), respondToIndustryMemoryBenchmarkSmokeCase(), scoreMemoryBenchmarkArtifact(), and isKnowledgeMemoryBenchmarkCase().
• Scores memory separately from RAG: current fact recall, stale-memory safety, event/source recall, actor recall, and stale rate.
• Adds calibration coverage showing a current-memory answer scores high while a stale-memory answer scores low.

Verification

• pnpm install --frozen-lockfile --prefer-offline
• pnpm exec tsc --noEmit
• pnpm exec vitest run tests/benchmarks.test.ts — 7 passed
• pnpm lint
• git diff --check
• pnpm exec vitest run — 266 passed, 12 skipped
• pnpm build
• File-backed proof: /tmp/agent-knowledge-memory-smoke-proof-final/knowledge-benchmark-report.json has 9 cases, 9 cells, 0 failures, mean score 1.0, 8 memory families, and 9 memory task kinds.

[tangletools]

✅ Auto-approved drewstone PR — fa4ce184

This PR was opened by the trusted drewstone account.
The full PR reviewer audit still runs separately and will publish findings if it detects issues.

_{tangletools · auto-approval · reason: drewstone_author · 2026-07-08T16:07:17Z}