feat(reporting): add researchReport executive summary layer

[drewstone]

Summary

researchReport composes the lower-level reporting primitives (summaryTable, paretoChart, gainHistogram, held-out GateDecisions, optional failureClusterView) into one structured launch-decision artifact. Verdicts are made on paired evidence — never on marginal means alone — and respect any held-out gate the caller passes through.

Decision rule

In order — first match wins:

1. Comparator itself → hold (baseline).
2. No comparator → hold if on the cost/quality Pareto frontier, else needs_more_data.
3. Held-out gate ≠ promote → reject (gate is necessary, not sufficient).
4. Paired N < 6 (RESEARCH_REPORT_HARD_PAIR_FLOOR) → needs_more_data.
5. ROPE configured AND paired-delta CI ⊂ ROPE → equivalent.
6. Paired-delta CI upper bound < 0 → reject.
7. Paired N < minPairs (default 20) → needs_more_data with MDE attached.
8. BH-adjusted q ≤ fdr AND paired CI excludes zero → promote.
9. Otherwise → hold.

What every candidate carries

• BH-FDR-adjusted Wilcoxon q-value, Cohen's d, marginal CI (from summaryTable).
• Bootstrap CI on the median paired delta (from gainHistogram).
• Bayesian-bootstrap-style Pr(Δ > 0) and Pr(Δ ∈ ROPE) on the mean paired delta (Rubin 1981 bootstrap-prior duality).
• Minimum detectable paired effect at the configured power / α via the new pairedMde primitive in power-analysis.
• On-Pareto-frontier flag (cost vs quality).
• Held-out gate label, when supplied.

What every report carries

• runFingerprint — SHA-256 over the canonicalised input run set (stable across input order).
• preregistrationHash — optional, links a signed HypothesisManifest so reviewers can verify the analysis was the preregistered one rather than post-hoc.
• methodology — structured assumptions / methods / alternatives / when-not-to-apply / citations, also embedded in the rendered markdown.
• Standalone methodology companion at docs/research-report-methodology.md.

Citations: Benjamini & Hochberg 1995; Wilcoxon 1945; Efron 1979; Rubin 1981; Kruschke 2018; Howard et al. 2021 (background reading on always-valid extensions).

API change

researchReport is now async — it uses Web Crypto via hashJson for the run fingerprint. No prior consumers exist (this PR is the first time the function ships).

Wired into src/index.ts (root barrel) and the @tangle-network/agent-eval/reporting subpath, plus RESEARCH_REPORT_HARD_PAIR_FLOOR and ResearchReportMethodology.

Adds a CLAUDE.md authorship directive (no AI-attribution trailers in this repo).

Test plan

• pnpm typecheck clean
• pnpm build produces dist/reporting.{js,d.ts} with the new exports
• pnpm test — 837 / 837 (11 dedicated researchReport cases covering: promote, needs-more-data + MDE, hard-floor below 6 pairs, ROPE → equivalent, gate override, fingerprint determinism + preregistration passthrough, failure-cluster surfacing — plus the existing summaryTable / paretoChart / gainHistogram coverage).
• Spot-check rendered markdown / HTML in a downstream consumer before tagging a release.