Spatial Postgres backend

[Solcany]

1. Spatial Postgres backend — the data layer moved to Postgres + PostGIS (replacing the earlier binary-file approach)
   and now ships as a custom image bundling the pg_roaringbitmap extension.

2. Roaring-bitmap precompute rollup — heatmap and histogram counts are precomputed into a cell_features index at
   rebuild time, cutting per-request query time from ~13s to under a second while staying exact at any resolution or
   filter.

3. Multi-dataset Linked-Data ingestion — ingestion loads streets, neighbourhoods, districts, and addresses from
   Adamlink TTL, plus the Beeldbank, Joods Monument, and Delpher record datasets, each linked to canonical places.

4. Point/line/polygon place geometry — places gained a dedicated geometry model supporting lines and polygons (not
   just points), with historical names and validity dates.

5. Spatially accurate heatmap — lines and polygons are rasterized into grid cells via PostGIS, and cells are
   reprojected from the Dutch RD grid to WGS84 with proj4 so they're square and grid-aligned.

6. Interactive filtering and browsing — the map gained filtering by dataset, record type, and geometry/place type, a
   time slider, and a side panel for browsing a cell's features.

7. Production deployment pipeline — Dockerized self-hosted and existing-Postgres setups, CI workflows that build and
   publish images to GHCR, and schema/ETL run from the image itself.

8. Non-blocking frontend rendering — heatmap and histogram are fetched client-side and cancellably so the map shell
   renders instantly, with sparse-heatmap rendering to keep payloads small.

9. Log-scale density normalization — heatmap color and histogram height are driven by spatial and temporal
   frequencies normalized on logarithmic scales, so relative density reads correctly across the city and timeline.

10. Schema consolidation — datasets and organisations tables, an Entity JSONB column, and URI-namespaced feature IDs
    for cleaner, deduplicated data.

11. Caching and connection resilience — a TTL query cache and pooled, timeout-guarded database connections.

12. Dynamic timeline — time slices are generated from the data's actual date range at index time rather than
    hardcoded.

13. Basemap provider switch — tiles moved from MapTiler to OpenFreeMap, dropping the API-key dependency.

14. Documentation — a README (architecture, ingestion, indexing, deployment) plus an mdsvex-rendered
    About page, LICENSE, and CITATION.