feat(benchmarks): add knowledge benchmark suite

[drewstone]

What changed

• Adds a reusable @tangle-network/agent-knowledge/benchmarks entrypoint for retrieval, RAG answer quality, hallucination, and KB-improvement benchmark cases.
• Declares the industry benchmark families we care about: BEIR, MTEB retrieval, MS MARCO/TREC DL, MIRACL, LoTTE, BRIGHT, CRAG, HotpotQA/KILT, RAGTruth/FaithBench, and first-party KB-improvement suites.
• Wraps the existing campaign runner so benchmark runs inherit resumability, traces, cost, latency, split summaries, and persisted JSON/Markdown reports.
• Adds qrels/JSONL parsers, normalized scoring dimensions, docs, and smoke tests.

Verification

• pnpm exec tsc --noEmit
• pnpm exec vitest run
• pnpm build
• pnpm lint
• git diff --check
• git merge-tree --write-tree origin/main HEAD
• Proof artifact written at /tmp/agent-knowledge-benchmark-proof-clean/knowledge-benchmark-report.json and .md with 2 cells, 0 failures, mean score 0.5.

Agent-eval reusable benchmark primitives landed in tangle-network/agent-eval#326.

[tangletools]

✅ Auto-approved drewstone PR — 2166fd1d

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 · 2026-07-08T01:18:55Z}