feat: CDC/SCD example notebook + CI notebook wiring

[mingjerli]

Summary

• Adds examples/cdc_scd_pipeline.ipynb — an end-to-end Debezium CDC → SCD Type 2 → fact → mart pipeline (6 statements, 4 layers, Databricks dialect) that stress-tests clgraph against realistic production shapes and showcases all 10 column-lineage features recently closed by the CDC/SCD gap program (gaps 1–10).
• Wires python run_all_notebooks.py --skip-llm into CI as a new notebooks job in .github/workflows/ci.yml, so every example notebook is guaranteed to execute cleanly on every PR (was previously unenforced).
• Updates docs/superpowers/specs/2026-04-13-cdc-scd-pipeline-gaps-design.md — progress summary, gap table, fix list, deliverables, and acceptance criteria all reflect that gaps 1, 2, 8 are now closed (alongside 3, 4, 5, 6, 7, 9, 10). All 10 gaps are resolved.
• Polishes gap 4 & gap 7 design docs with accurate line references and edge-role naming (cross_query_self_ref). No behavior change.

What the notebook captures on the pipeline

Gap	What's demonstrated	Edges on this pipeline
1	Struct dot-access on CDC envelope (after.id, after.name, …)	5 struct edges, access_type=struct, nested_path
2	Dedup via subquery + WHERE rn = 1	1 qualify edge, qualify_function=ROW_NUMBER
3/10	MERGE WHEN MATCHED AND (…) condition columns	13 edges, merge_column_role=condition
4	Self-referencing target (MERGE then INSERT on dim_customer)	4 self-read columns, 4 cross-query self-ref edges, 28 prior-state-read edges
5	Literal-only outputs ('Y' AS is_active)	terminal nodes, no upstream edges
6	current_timestamp() function-only source	output nodes, no incoming edges
7	JOIN ON predicate columns incl. BETWEEN (point-in-time fact join)	5 predicate edges, is_join_predicate=True
8	WHERE clause columns	50 where_filter edges with where_condition metadata
9	MERGE ON literal-bound predicate (t.is_active = 'Y')	1 merge_match_filter edge

Total: 95 edges across 40 columns, 5 queries.

Test plan

• uv run pytest tests/ — 1562 passed, 41 skipped, 2 xfailed (unchanged from baseline)
• uv run python run_all_notebooks.py --skip-llm — 27/27 pass (including the new notebook)
• uv run ruff check examples/cdc_scd_pipeline.ipynb — clean
• uv run ruff format --check examples/cdc_scd_pipeline.ipynb — clean
• CI green on new notebooks job (verify after push)