refactor: dashboard ACB v5 — 2 fonti, -42% righe

[Gabrymi93]

Summary

Refactoring completo della dashboard: da 5 fonti dati a 2 (ACB + SO direct), con pagine riscritte e -42% righe.

Architettura

Prima:

lab-dashboard → dataset-incubator/registry.json
lab-dashboard → source-observatory (6 file)
lab-dashboard → data-explorer/themes.json.py
lab-dashboard → dataciviclab/analisi/ (GitHub API)
lab-dashboard → GitHub GraphQL (discussions)
lab-dashboard → GCS parquet
= 5 fonti, ~15 fetch, ~1900 righe

Dopo:

lab-dashboard → agent-context-builder/context (2 JSON)
lab-dashboard → source-observatory (5 file)
lab-dashboard → GCS parquet
= 2 fonti, ~10 fetch, ~1100 righe

Changes

sources.py (480 → 258 righe, -46%):

• ACB loaders: load_catalog(), load_signals(), load_radar(), load_explorer_datasets(), load_discussion_counts(), load_discussions(), load_workspace_triage()
• SO direct: load_radar_history(), load_sources_registry(), load_sources_dashboard(), load_source_report(), load_catalog_signals(), load_inventory_report(), load_check_coverage()
• Eliminati: GraphQL, GitHub API, themes.json.py

00 Vista Insieme (209 → 150 righe):

• KPI compatta: Radar, Dataset, Pipeline, PR/Issues/Discussions
• Barra radar segmentata (non bar chart)
• Fonti non healthy in expander
• Top 15 fonti con bar chart + tabella compatta
• Nessun expander per dataset

01 Dataset Explorer (110 → 165 righe):

• Raggruppamento per source_id (bar chart)
• Filtri: fonte, stage, categoria, ricerca
• Tabella compatta + espansione per dettaglio
• Copertura anni + verifica parquet

02 Registry / Repo (254 → 109 righe):

• KPI aggregati: repo attivi, dataset totali, mart, segnali
• Tabella repo: dataset, mart, segnali, ok/warn/error
• Bar chart dataset per repo
• Repo senza registry

09 Query SQL (334 → 56 righe, -83%):

• Delega a lab_connectors.duckdb.sql_page.render_sql_query()
• Adapter: ACB → Registry object

pyproject.toml: [project] completo con deps (ADR-001 §8)
requirements.txt: +lab-connectors[duckdb] + agent-context-builder
tests: 46/46 passano, ruff clean

Test

• 46/46 test passano
• ruff check clean
• Nessun import inutilizzato