feat(demo): LakeTS Live streaming demo on Lakebase Autoscaling

[taran-dbx]

Summary

Brings the standalone lakets-live-demo into the repo as demo/live/, adapted to current LakeTS capabilities, plus a published step-by-step setup guide. A living end-to-end demo: synthetic ticks stream into a Lakebase Autoscaling project while serverless jobs drive partitioning, DAG-ordered RollUp refresh, CDF-gated tiering, and retention; Lakebase CDF replicates to Unity Catalog.

What's adapted to current capabilities

• Autoscale M2M OAuth + psycopg3 — stream_ticks resolves the project's primary read-write endpoint via w.postgres and mints a per-connection OAuth credential (mirrors lakebase_utils.py). No static passwords.
• Reuses the shipped maintenance jobs — the demo bundle's partition/rollup/tiering/retention tasks point at databricks/workflows/*.py (single source of truth, no drift). The demo adds only stream_ticks.
• Cold tier via lakets.enable_sync (Lakebase CDF; shadow in lakets_cdf) instead of a manual UI sync.
• create_rollup without the removed p_refresh_mode arg; tiering_job (CDF durability gate) instead of a Spark compression step.

Repo fix (folded in)

rollup_refresh.py now refreshes in DAG order via lakets.refresh_rollup_cascade() instead of an alphabetical refresh_rollup() loop (which refreshed 1day before 1hour before 1min). New guard test test_refreshes_in_dag_order.

Convention change

• demo/ is no longer gitignored (and the any-depth grafana/ pattern that would have re-ignored demo/live/grafana/ is removed; .env.example un-ignored). Old local demos were archived out of the repo; only this demo ships.

Docs

New Docusaurus guide Guides → Demos → Live Demo (guides/live-demo): prerequisites (incl. CDF), install, setup.sql, bundle deploy (dev + prod SP), mid-demo knobs, Grafana, teardown.

Test plan

• All workflow files compile; tests/test_python_patterns.py passes (14, incl. new DAG-order guard)
• Docusaurus build passes (no broken links/anchors)
• bundle validate -t dev OK against the autoscale project; setup.sql cross-checked against current function signatures
• Full live run (deploy + stream + watch tiering) on lakets-tiering-test — next step