fix(pipeline): eurostat filter bugs + supervisor uses sqlmesh plan

- nrg_pc_203: add missing unit=KWH filter (API returns 2 units)
- lc_lci_lev: fix currency→unit filter dimension name
- supervisor: use `sqlmesh plan prod --auto-apply` instead of
  `sqlmesh run` so new/changed models are detected automatically

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

CHANGELOG.md

@@ -8,6 +8,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ### Fixed
 - **Pipeline diagnostic script** (`scripts/check_pipeline.py`) — handle DuckDB catalog naming quirk where `lakehouse.duckdb` uses catalog `lakehouse` instead of `local`, causing SQLMesh logical views to break. Script now auto-detects the catalog via `USE`, and falls back to querying physical tables (`sqlmesh__<schema>.<table>__<hash>`) when views fail.
+- **Eurostat gas prices extractor** — `nrg_pc_203` filter missing `unit` dimension (API returns both KWH and GJ_GCV); now filters to `KWH`.
+- **Eurostat labour costs extractor** — `lc_lci_lev` used non-existent `currency` filter dimension; corrected to `unit: EUR`.
+- **Supervisor transform step** — changed `sqlmesh run` to `sqlmesh plan prod --auto-apply` so new/modified models are detected and applied automatically.
 
 ### Added
 - **Pipeline diagnostic script** (`scripts/check_pipeline.py`) — read-only script that reports row counts at every layer of the pricing pipeline (staging → foundation → serving), date range analysis, HAVING filter impact, and join coverage. Run on prod to diagnose empty serving tables.