ci: lint/type/test floor + dataset & eval-baseline gates (Milestone A)

[helebest]

What

Closes the CI/CD gap in dikw-data so it has the same deterministic floor as dikw-core, before Phase 1 dataset construction begins. This is Milestone A of the eval-plan Phase 0→1 work (see docs/dikw-eval-plan.md); the public-anchor calibration (scifact/cmteb) follows as Milestone B.

Changes

• pyproject.toml — ruff + mypy (strict) config mirroring dikw-core. Ignores RUF001/2/3 (ambiguous-unicode false positives on this bilingual zh/en codebase's embedded CJK text) and scope-ignores E702 to the one procedural-pictogram generator.
• .pre-commit-config.yaml — local ruff + mypy hooks (uv run), matching CI. Install with uv run pre-commit install.
• .github/workflows/ci.yml — uv sync → ruff → mypy src → pytest → validate every dataset (shape gate, $0, no provider keys). Matrix 3.12 / 3.13.
• .github/workflows/eval-gate.yml + tools/check_baselines.py — a datasets/** change must land a new dated reports/BASELINES.md entry naming a retrieval metric; override with the no-baseline-needed label. The pure check is unit-tested (tests/test_check_baselines.py).
• .gitignore — track reports/BASELINES.md (the baseline log) while keeping per-run artifacts ignored; ignore .impeccable/.
• Applied ruff autofixes (import sorting, unused-import / whitespace cleanup) across scripts/, src/, web/, tests/ so the existing tree passes the new gate. Fixed 3 real lint findings (RUF005 in run_eval.py, dead code in generate_queries_local.py) and 2 mypy findings (unused ignore; int**int Any-widening in llm_client.py).

Verification (local)

• uv run ruff check . → clean
• uv run mypy src → clean
• uv run pytest → 50 passed
• scripts/validate_dataset.py over all 3 datasets → valid
• eval-gate plumbing tested end-to-end: non-dataset change → no-op pass; dataset change without a baseline entry → fail (exit 1); with a proper entry → pass.

🤖 Generated with Claude Code