feat: add cb health-check command (extract → eval → gap report)

[rajnavakoti]

Summary

Packages the full pipeline into a single command — cb health-check ./docs/ runs extract → eval → gap report and prints a knowledge-health summary. No prior cb init required. This is the "free health check" flow for conference/design-partner outreach.

Exit code is a deployment gate: 0 if CLEAN coverage ≥ threshold, else 1.

What it does

1. Extract entities from the docs folder (synthesizes a minimal seed if none given)
2. Generate eval questions (seed + doc + persona) and run coverage eval
3. Produce an HTML gap report
4. Summarize: entity count, coverage %, per-persona breakdown, top-5 gaps (by severity), cost estimate

Changes

• context_blocks/cli.py — new health-check command (mostly orchestration of existing pieces)
• context_blocks/retrieval/evals.py — extracted a shared run_coverage_eval() helper (KB load + embed + retrieval + coverage report), so the eval flow is reusable
• tests/unit/test_health_check.py — new

Test plan

• Pass above threshold → exit 0, HTML report written
• Below threshold → exit 1
• Missing docs dir → exit 1 with error
• Synthesizes a seed when none provided
• Summary shows entities + coverage
• Full unit suite: 205 passed, no regressions

Follow-ups (not in this PR)

• Refactor the existing cb eval command to also use run_coverage_eval() (DRY) — kept separate here to keep the change low-risk.
• Extraction cost isn't tracked upstream, so the cost line reflects the eval estimate only (labelled as such).
• PDF report generation is a separate issue (per feat: health check packaging — cb health-check command #26 scope).

Closes #26

🤖 Generated with Claude Code