Show independent model evidence without promotion

[mattfaltyn]

Summary

• Remove automatic model promotion, champion selection, and rollback. EWMA remains the default; explicit user selection controls compact odds, sort, and hypothetical risk.
• Show all five physical and two market model results with exact unavailable reasons. Add independent history blocks, bounded Monte Carlo sampling error, fit timing, model spread, and separate prospective versus retrospective accuracy evidence.
• Restrict displayed accuracy and coverage to current model versions and label the conditional coverage cohort. Preserve existing DuckDB data and legacy champion files without using them for live selection.

Validation

• scripts/verify-fast: 397 backend and 150 frontend tests passed.
• scripts/verify: Ruff, 397 backend tests, npm audit (0 vulnerabilities), 150 frontend tests, lint, API generation check, build, and 27 Playwright tests passed.
• Independent R2 quant review: PASS.

Pad: PLAN-94; WORK-95, WORK-96, WORK-97.