Design: OSM way/relation ingestion and sub-place hierarchy

[koinsaari]

Goal

Design and implement ingestion of OSM ways and relations (polygons) — large venues like malls, hospitals, universities, parks — and wire them into the parent_id hierarchy so enclosed OSM nodes become sub-places.

Tracking under #10.

Context

Current ingestion only processes OSM nodes. Ways and relations are skipped entirely. This means:

• A mall exists in OSM as a way (polygon), not a node — it is absent from places today.
• Individual shops inside that mall exist as nodes and are ingested — but with parent_id = NULL, as if they were standalone.
• The 3-level rank system (1=Landmark, 2=Establishment, 3=Feature) is derived purely from category + tags, with no awareness of spatial containment.

Decisions already made

• Schema is ready: geometry JSONB, parent_id UUID REFERENCES places(id) both exist.
• Way ingestion auto-assigns parent_id via spatial containment — ingestion does the spatial join, not the API.
• This is additive: existing Wheelmap external_ids attached to node-places are unaffected when parent_id is later set on those nodes.
• The retroactive containment sweep (assign parent_id to nodes already in DB that fall inside a newly ingested way) must be idempotent.

Open questions

• Rank for contained nodes: should a shop node spatially inside a mall way be demoted from rank 2 → 3? Or does rank stay category-derived and clients filter by parent_id? Decide before implementing.
• Geometry representation: ways are polygons — store as GeoJSON in geometry JSONB, or add a PostGIS GEOMETRY column? The existing geometry JSONB column was added speculatively; evaluate whether PostGIS native type is worth the migration cost.
• Relation handling: OSM relations (multipolygons, e.g. large parks) — in scope for this issue or a follow-up?
• Deletion: when a way is deleted in an OSM diff, do its former child nodes get parent_id cleared, or stay associated?
• Rank constraint: current CHECK (rank IN (1, 2, 3)) covers the node world. If contained nodes get a 4th rank level, migration needed — trivial pre-production, needs a plan post-production.

Out of scope

• Non-OSM sources for polygon data.
• Moderation or review UI for auto-assigned parent relationships.
• Manual parent_id override via API (already works today).

Acceptance

• OSM way polygons ingested as places rows with geometry populated.
• Existing nodes spatially inside ingested ways have parent_id set (retroactive sweep + ongoing).
• Rank behavior for contained nodes is explicitly decided and documented.
• At least one integration test covering a small way fixture with enclosed nodes.