refactor(web): derive content from repository sources - #142

Summary

• derive every website benchmark value, comparison, chart, table, date, and accessible label from the generated benchmark results artifact
• validate the benchmark schema and provenance at build time with no stale fallback
• derive rendered and copyable website documentation from README and the generated command reference
• add the provenance validator to web lint and record completion in the ADR and backlog

Safety and correctness

• keeps repository file access in server-rendered modules only
• bounds and validates every source file before parsing
• rejects missing cases, duplicate cases, invalid metrics, unsupported schema versions, and non-task-aware benchmark data
• derives comparison direction and headline claims so future regressions cannot be labeled as improvements
• adds no runtime endpoint or dependency

Validation

• pinned Prettier 3.6.2 check on every changed file
• pnpm --filter @headless/web lint
• pnpm --filter @headless/web build
• rendered /, /docs, and /docs/markdown verification
• pnpm test
• pnpm test:runtime
• pnpm test:npm
• shell syntax and QA evidence checksum checks
• ./apps/headless/build.sh
• pnpm test:e2e:linux

Closes #48