Support generic supplied document injection

[lilyjge]

Summary

• add generic supplied document bundles for benchmark runs via --supplied-doc-bundle / SUPPLIED_DOC_BUNDLE
• inject grouped supplied evidence into the agent prompt and record metadata.supplied_docids
• count supplied docids in surfaced, previewed, and agent-visible retrieval views
• document the bundle JSONL format and wire the high-level query-set launcher to forward the option

Verification

• npm run typecheck
• npm run format:check
• npm run lint
• git diff --check
• npx tsx --test tests/benchmark_query_set_launch.test.ts tests/launcher_wrappers.test.ts tests/run_pi_benchmark_pi_search_contract.test.ts

Note

• Full npm test is 229/230 locally; the unrelated failing test expects missing local fixture data/msmarco-v1-passage/queries/dl19.tsv in compare_bm25_runs.test.ts.

[lilyjge]

@lingwei-gu for context