perf: report review context budget ledger

[stainlu]

Summary

• Add a deterministic review-context budget ledger to generated ClawSweeper reports.
• Record per-section prompt JSON chars plus retained/hydrated/total counts and truncation state.
• Keep the ledger report-only so review prompts and model behavior do not change.
• Include dashboard/ in CI sparse checkout so existing dashboard worker tests can run on PRs.
• Add unit coverage for ledger ordering, section accounting, and rendered budget text.

Why

ClawSweeper is now doing real context compaction, but maintainers still need durable per-run numbers to see which prompt sections dominate cost. This turns review reports into a cheap measurement surface before we make more aggressive token/time optimizations.

The CI sparse checkout already included test/, but omitted tracked dashboard/, so test/dashboard-worker.test.ts failed to import dashboard/worker.js on GitHub. This branch includes the minimal workflow fix needed for CI parity.

Validation

• pnpm run build
• pnpm run format:check
• pnpm run test:unit
• git diff --check

Note

Local runs are on Node 22 and show the repo engine warning (>=24). I kept validation to the changed source/unit surface locally and expect GitHub CI to run the full Node 24 pnpm check lane.