feat: full-text search + complete dpe

[il-dat]

What & why

Two capabilities in one change.

1. Full-text search overhaul (SPA). The bundled SPA's search becomes a real full-text search instead of a name/column-only lookup. A query now indexes the whole searchable surface of every node — name, columns and column descriptions, tags, description, warehouse relation, package, called macros, and the raw + compiled SQL — and each non-name hit carries a match-reason snippet (mkdocs-material style): a labelled chip (Column, SQL, Tag, …) plus an excerpt with the matched terms highlighted, so you can see why a result is there. Two inline operators (type:<resource_type> and label:/name:) narrow without leaving the box, and the dropdown is fully keyboard-operable (↑/↓ rove, Enter follows, Esc closes) and wired as an ARIA combobox/listbox with a live region for screen readers.

2. Complete dbt-project-evaluator (DPE) health-rule parity. dbdocs health rules now implement all 29 published DPE rules across the six dimensions:

• Removed the non-DPE orphan_models rule — it had no DPE counterpart, so its docs_url linked to a dead anchor.
• Added the 5 missing DPE rules, each with a valid docs anchor and unit tests:
  • downstream_models_dependent_on_source, hard_coded_references (modeling)
  • missing_source_freshness (testing)
  • documentation_coverage (documentation)
  • test_directories (structure)
• hard_coded_references parses each model's raw SQL with sqlglot (ref()/source() jinja rewritten to numbered sentinels; CTEs excluded via scope analysis; fail-soft on unparseable SQL — one model never sinks the pass).
• Fixed 3 dead docs anchors via a new name→anchor override map in base.py (the rule name doesn't always kebab to the DPE heading): too_many_joins, staging_dependent_on_marts_or_intermediate, and staging_dependent_on_staging.
• Added a weekly watcher (.github/workflows/dpe-rules-watch.yml + .github/scripts/check_dpe_rules.py) that scrapes the DPE rules pages, diffs them against the live rule registry, and opens (or updates) a single feat: issue when DPE publishes a rule dbdocs doesn't have yet. The watcher is what surfaced the third dead anchor.

Type of change

• 🐞 Bug fix (non-breaking change that fixes an issue)
• ✨ Feature (non-breaking change that adds capability)
• 💥 Breaking change (existing behavior changes)
• 🧹 Refactor / internal (no user-facing change)
• 📖 Docs only

Note: removing orphan_models is technically a behavior change to the Health section, but it never matched a DPE rule and only ever produced a finding with a broken docs link.

Area

• CLI (generate / serve / deploy)
• extract/ (nodes / erd / graph / column_lineage / health)
• site/ (data dict / builder / 3-tier SPA / deploy)
• frontend/ (React Flow graph bundle)
• Config (dbdocs.yml) / packaging / CI / docs

How to test

task generate && task serve   # then use the search box:
#   count_food_items   → a result with a "Column" snippet (matched a column, not a name)
#   type:seed          → lists every seed, nothing else
#   type:model orders  → only models matching "orders"
#   label:stg          → names containing "stg", ignoring SQL-body mentions
#   type:bogus         → "No matches." cue (not a silent empty dropdown)
#   ↑ / ↓ / Enter / Esc → keyboard nav of the dropdown

task test            # 100% coverage (331 tests), incl. the new DPE rules + watcher logic
task frontend:e2e    # the new Playwright search specs (Node + a real demo build)

# Verify full DPE parity + that every rule's docs anchor resolves (hits the live DPE pages):
uv run python .github/scripts/check_dpe_rules.py
#   → "dbdocs implements every published dbt-project-evaluator rule."

Checklist

• task lint passes (ruff format --check + ruff check).
• task test passes at 100% coverage (331 tests, 100.00% total).
• I followed the load-bearing patterns in .claude/design_patterns.md. Extended the health-rule registry (Always-built Health-Check pattern: one module per dimension, register_rule/docs_url/finding public API) and the 3-tier data→service→ui SPA seam (service.js DOM-free, ui.js the only DOM toucher). No pattern added/removed — the doc describes the rule package generically, so no edit needed.
• Data-dict / SPA changes keep producer ↔ consumer in sync and don't re-inline the payload. Search reads existing data-dict fields only — no Python builder change, payload stays the external gzip; new logic lives in service.js (pure) + ui.js (DOM). Health findings keep the same {rule, node, node_type, message, docs_url} shape.
• Graph-UI changes rebuilt the committed bundle — n/a: no frontend/src/** change; only frontend/test/e2e/spa.spec.ts was touched.
• New files under dbdocs/site/bundle/** — n/a: no new bundle files (bundle assets edited in place; the dbdocs/site/bundle/**/* glob already covers them). New files are under .github/ + tests/, not shipped in the wheel.
• Docs / dbdocs.yml.example updated. dbdocs.yml.example + docs/dbdocs-demo.yml list all 29 rules + the new documentation_coverage threshold; the search guide (docs/nav/guide/search.md) is wired into mkdocs.yml; README/docs/index.md feature line refreshed.

Screenshots / notes

• Site change — a before/after of the search dropdown belongs here, showing the new match-reason snippet (labelled chip + highlighted excerpt).
• Search behavior is verified by the new Playwright specs in frontend/test/e2e/spa.spec.ts, which run in the independent CI e2e job (outside the coverage gate, since pytest doesn't exercise the SPA JS).
• The DPE watcher runs weekly (Mondays 06:00 UTC) and on manual dispatch; it dedupes on a stable issue title so it updates rather than spams.