refactor(evaluation): remove legacy loops and reduce gate overhead

[drewstone]

Problem

Supervisor-run still published and read the retired .loops control-plane layout, including legacy terminal status fields. Paired binary and pass/fail gates also paid for unnecessary repeated numerical work.

Change

• Delete the .loops reader, compatibility exports, legacy status sources, and fallback wording.
• Make reader.ts the Runtime-only filesystem entrypoint while preserving the pure source analyzer and adapter seam.
• Return unavailable when only state.status or result.sup_status exists.
• Reduce score-interval bisection to the precision supported by the calculation.
• Use the exact binomial shortcut for equal-magnitude Wilcoxon differences.
• Refresh the public analyst implementation pin and the supervisor-run API census counts.

Validation

• pnpm test — 409 files passed; 6,112 tests passed; 2 files and 3 tests skipped.
• pnpm build
• pnpm verify:package
• pnpm typecheck
• pnpm typecheck:examples
• pnpm typecheck:scripts
• pnpm lint
• pnpm vitest run src/supervisor-run — 164 tests passed.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.