feat: Complete Phase 5 dashboard implementation

Implement full 5-tab Toronto Neighbourhood Dashboard with real data
connectivity:

Dashboard Structure:
- Overview tab with livability scores and rankings
- Housing tab with affordability metrics
- Safety tab with crime statistics
- Demographics tab with population/income data
- Amenities tab with parks, schools, transit

Figure Factories (portfolio_app/figures/):
- bar_charts.py: ranking, stacked, horizontal bars
- scatter.py: scatter plots, bubble charts
- radar.py: spider/radar charts
- demographics.py: donut, age pyramid, income distribution

Service Layer (portfolio_app/toronto/services/):
- neighbourhood_service.py: queries dbt marts for all tab data
- geometry_service.py: generates GeoJSON from PostGIS
- Graceful error handling when database unavailable

Callbacks (portfolio_app/pages/toronto/callbacks/):
- map_callbacks.py: choropleth updates, map click handling
- chart_callbacks.py: supporting chart updates
- selection_callbacks.py: dropdown handlers, KPI updates

Data Pipeline (scripts/data/):
- load_toronto_data.py: orchestration script with CLI flags

Lessons Learned:
- Graceful error handling in service layers
- Modular callback structure for multi-tab dashboards
- Figure factory pattern for reusable charts

Closes: #64, #65, #66, #67, #68, #69, #70

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

docs/project-lessons-learned/sprint-9-10-graceful-error-handling.md

@@ -0,0 +1,34 @@
+# Sprint 9-10 - Graceful Error Handling in Service Layers
+
+## Context
+Building the Toronto Neighbourhood Dashboard with a service layer that queries PostgreSQL/PostGIS dbt marts to provide data to Dash callbacks.
+
+## Problem
+Initial service layer implementation let database connection errors propagate as unhandled exceptions. When the PostGIS Docker container was unavailable (common on ARM64 systems where the x86_64 image fails), the entire dashboard would crash instead of gracefully degrading.
+
+## Solution
+Wrapped database queries in try/except blocks to return empty DataFrames/lists/dicts when the database is unavailable:
+
+```python
+def _execute_query(sql: str, params: dict | None = None) -> pd.DataFrame:
+    try:
+        engine = get_engine()
+        with engine.connect() as conn:
+            return pd.read_sql(text(sql), conn, params=params)
+    except Exception:
+        return pd.DataFrame()
+```
+
+This allows:
+1. Dashboard to load and display empty states
+2. Development/testing without running database
+3. Graceful degradation in production
+
+## Prevention
+- **Always design service layers with graceful degradation** - assume external dependencies can fail
+- **Return empty collections, not exceptions** - let UI components handle empty states
+- **Test without database** - verify the app doesn't crash when DB is unavailable
+- **Consider ARM64 compatibility** - PostGIS images may not support all platforms
+
+## Tags
+python, postgresql, service-layer, error-handling, dash, graceful-degradation, arm64