feat(analysis): exploratory ranks + multi-session claim tooling

[leo-gan]

Summary

• Multi-way effect-vs-fastest now uses median reference, Mann–Whitney tests, and within-group Holm adjustment; Results pages label ranks as exploratory (prefer pairwise A/B for confirmatory checks).
• Sidecars gain machine_id / L1 claim hint; --multi-session aggregates rank stability and L1/L2/L3 claim levels; soft check-research-env.sh for research hosts.
• Student docs: claims ladder (CLAIMS_AND_REPLICATION.md), methodology sections, Wikipedia links for statistical terms; strip internal ticket tags from public docs.

Validation

• Tests: analysis 83 pass; python 29 pass; javascript 9 pass; rust 13 pass; cpp roundtrips OK. Java/mvn unavailable (warn only).
• Benchmarks: skipped (no benchmark-runner language changes; analysis/docs only).
• Error CSVs: N/A (no rebench).
• Analysis: no forced full-matrix Results regen (banners apply on next analyze-benchmarks).
• Dashboard: sync-data.py verified latest stems present; no commit of timestamp-only gzip churn.

Notes

• Default Results remain L1 single-session / single-host; multi-session is opt-in CLI.
• Older CSVs without machine_id in sidecars report L2_multi_session_host_unknown when aggregating ≥3 sessions.
• Example: analyze-benchmarks --multi-session python:STEM1 --multi-session python:STEM2 --multi-session python:STEM3

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.