v0.5.6 — DB query optimization and GDPR atomicity

Performance

• Prisma query payload reduction. Five findUnique calls without select now fetch only the columns actually consumed: badge-update fetches totalCount only (was all 15+ columns), map-image fetches 4 badge columns, stargazer-cache GET fetches updatedAt for the fallback path and explicit 5-column select for the full row. Reduces data transferred from Neon on every scan completion and every map image generation.
• organic-score-stats SQL aggregate. The admin stats endpoint was loading all badge_cache rows into Node.js memory for in-memory aggregation. Replaced with two parallel $queryRaw GROUP BY queries (per-tier counts + per-bucket distribution via width_bucket). Memory footprint is now O(1) regardless of table size.

Internal

• import-geocache bulk upsert. N+1 pattern replaced with a single $queryRaw UNNEST INSERT ... ON CONFLICT DO UPDATE per batch of 500. Same pattern as bulkUpsertUsers in user-cache.ts. Admin dev-only route, blocked in production.
• GDPR deletion atomicity. delete-user route wraps starEvent.deleteMany + gitHubUser.delete in prisma.$transaction([...]). Prevents partial deletion if the process crashes between the two operations.