feat(postcodes): backfill state_id on 19,272 rows across 7 countries (#1039)

[dr5hn]

Summary

Backfills state_id (and state_code) on 19,272 existing postcode rows that previously shipped with state_id: null. No new external data — uses CSC's own contributions/cities/<iso2>.json as a city → state_id lookup table.

iso2	Country	Total rows	Backfilled	Match %
SE	Sweden	16,392	13,162	80%
ZA	South Africa	5,685	2,906	51%
NL	Netherlands	4,072	2,463	60%
RS	Serbia	1,170	343	29%
SI	Slovenia	522	271	51%
GB	UK	124	98	79%
VN	Vietnam	63	29	46%
	Total	27,028	19,272	71%

Why this is a strict improvement

• Rows that do match are now fully FK-resolved.
• Rows that don't match remain state_id: null — the same as before.
• Names that map to multiple distinct states (e.g., NL's Rijswijk exists in both Noord-Brabant and Zuid-Holland) are deliberately not matched, since postcode data alone can't disambiguate. 15 ambiguous names were dropped from the NL lookup, 24 from SE, 29 from GB.

Why these 7 countries (and not others)

Audit of all 0%-FK postcode files vs cities.json:

Country	Verdict	Reason
LV	skipped	source ships codes without locality_name (no field to match)
MT	skipped	"localities" are street names, not cities
MU/GR/KE/NP	skipped	localities are post-office names; <15% match

Implementation

bin/scripts/sync/backfill_postcode_state_fk.py — single pass, idempotent, supports --dry-run and --only=<iso2[,iso2...]>. NFKD-fold names for diacritic-insensitive matching. Skips rows whose state_id is already set.

The diff size is large because the existing files were re-emitted with consistent JSON formatting (existing id, created_at, updated_at, flag fields preserved verbatim; only state_id and state_code mutated where matchable).

🤖 Generated with Claude Code